Data generation system and control method thereof

The data generation system automatically generates domain-specific and user-preference-based data using AI, addressing the challenges of time and security in existing methods, enhancing efficiency and accuracy in large-scale language model training.

KR103046532B1Active Publication Date: 2026-11-11LG MANAGEMENT DEV INST CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
KR1020250174667
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Priority Date
2024-11-20
Filing Date
2025-11-18
Publication Date
2026-11-11
Estimated Expiration
2045-11-18

AI Technical Summary

Technical Problem

Existing technologies face challenges in rapidly and securely obtaining high-quality, domain-specific data and data reflecting user preferences for training large-scale language models, which requires substantial time and manpower, and involves security risks with external expert involvement.

Method used

A data generation system and control method using artificial intelligence to automatically generate domain-specific question-response pairs and user preference-based responses, constructing data sets, and fine-tuning target models for customized training.

Benefits of technology

Enables efficient, user-customized, and high-quality data generation with minimal manpower and time, improving response accuracy and performance while ensuring security and automation from data generation to model training.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure R1020250174667_ABST
    Figure R1020250174667_ABST
Patent Text Reader

Abstract

The present invention relates to a data generation system and a control method thereof, and provides a data generation system and a control method thereof that automatically generates domain-specific data and data reflecting user preferences using an artificial intelligence model.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a data generation system and a control method thereof, and provides a data generation system and a control method thereof that automatically generates domain-specific data and data reflecting user preferences using an artificial intelligence model. Background Technology

[0002] The dictionary definition of artificial intelligence is a technology that realizes human learning, reasoning, perception, and natural language understanding abilities through computer programs. This artificial intelligence has achieved rapid development through deep learning.

[0003] In particular, driven by the advancement of artificial intelligence, various language models have been developed. These models have reached a level where they not only recognize text and understand its meaning but also extract and classify information from vast amounts of text-based data, such as documents, and even generate text directly.

[0004] These language models are actively utilized in various fields and exist in diverse areas where text-based tasks can be performed, such as search engines, document creation (e.g., resume writing, report writing, posting, etc.), free conversation on various topics, data parsing from given text (e.g., data summarization, classification, etc.), provision of expertise, programming, and converting given sentences into sentences of an appropriate style.

[0005] In this regard, Large Language Models (LLMs) have recently emerged, which understand and generate human language through prior training on vast amounts of text data. Unlike traditional chatbots, which are manually built and provide only limited answers, LLMs are demonstrating innovation in the artificial intelligence market by showcasing technological capabilities that allow them to communicate naturally, almost like humans, and provide fast and accurate information.

[0006] Consequently, active research is currently being conducted on methods to generate data used for training and evaluating large-scale language models in order to utilize them across various fields. However, despite these ongoing studies, rapidly and securely obtaining high-quality data specialized for specific domains (or fields) or data that reflects user preferences remains a significant challenge. For instance, securing sufficient data for training during model development requires substantial time and manpower, and utilizing external experts to obtain high-quality data may lead to security issues (or risks) involving the leakage of core data.

[0007] Therefore, to solve these problems, there is a need for a platform that can quickly and securely secure high-quality data specialized for specific domains or data reflecting user preferences based on artificial intelligence. (Patent Document 1) US 20180239811 A1 The problem to be solved

[0008] The present invention is intended to provide a data generation system capable of automatically generating high-quality data based on artificial intelligence and a method for controlling the same.

[0009] More specifically, the present invention aims to solve the problem of data scarcity and to provide a data generation system and a control method thereof that can automatically generate domain-specific data and data reflecting user preferences.

[0010] In addition, the present invention is intended to provide a data generation system capable of automating everything from high-quality data generation to model training, and a method for controlling the same.

[0011] Furthermore, the present invention aims to provide a data generation system capable of providing a user-customized model learned through high-quality data and a method for controlling the same. means of solving the problem

[0012] To solve the problem described above, a control method for a data generation system according to the present invention, which is performed by a computer, may include the steps of: receiving at least one user query for at least one document from a user terminal; at least one processor processing the at least one document and the at least one user query as input to at least one artificial intelligence model; generating at least one question-response pair satisfying at least one pre-set criterion based on the at least one document and the at least one user query in the at least one artificial intelligence model executed by the at least one processor; and constructing at least one data set using the at least one question-response pair.

[0013] In an embodiment, the at least one criterion set in advance includes at least one of a first criterion related to the generation of a domain-specific question-response pair and a second criterion related to the generation of a question-response pair reflecting a user's response preference, and in the step of generating the at least one question-response pair, the at least one question-response pair satisfying at least one of the first criterion and the second criterion is generated in the at least one artificial intelligence model executed by the at least one processor based on the at least one document and the at least one user query, and in the step of constructing the at least one data set, the at least one data set can be constructed using the at least one question-response pair satisfying at least one of the first criterion and the second criterion.

[0014] In an embodiment, the at least one question-answer pair comprises at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion, and in the step of generating the at least one question-answer pair, at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion can be generated from the at least one document and the at least one user query using the at least one artificial intelligence model performed by the at least one processor.

[0015] In an embodiment, the step of generating at least one question-answer pair may include at least one of the steps of generating at least one question-answer pair satisfying the first criterion from the at least one document and the at least one user query using the at least one artificial intelligence model performed by the at least one processor, and generating at least one question-answer pair satisfying the second criterion from the at least one document and the at least one user query using the at least one artificial intelligence model performed by the at least one processor.

[0016] In an embodiment, the step of generating at least one question-response pair satisfying the first criterion may include: the step of the at least one processor processing the at least one document and the at least one user query as inputs to at least one artificial intelligence model; the step of generating at least one augmented query by performing augmentation on the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; the step of generating at least one response to the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; and the step of generating at least one response to the at least one augmented query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor.

[0017] In an embodiment, in the step of generating at least one question-response pair satisfying the first criterion, at least one question-response pair satisfying the first criterion can be generated by using at least one of the at least one user query and the at least one response to the at least one user query and the at least one augmented query and the at least one response to the at least one augmented query.

[0018] In an embodiment, at least one question-response pair satisfying the first criterion may include at least one of at least one question-response pair configured to have at least one user query and at least one response to the at least one user query, and at least one question-response pair configured to have at least one augmented query and at least one response to the at least one augmented query.

[0019] In an embodiment, the step of generating at least one question-response pair satisfying the first criterion may be the step of generating at least one question-response pair specialized in at least one domain specified from at least one of the at least one document and at least one user query.

[0020] In an embodiment, at least one question-response pair satisfying the first criterion may be configured such that at least one question specialized in the at least one domain and at least one response to the question form a pair.

[0021] In an embodiment, the step of generating at least one question-response pair satisfying the second criterion may include: the step of the at least one processor processing the at least one document and the at least one user query as inputs to at least one artificial intelligence model; the step of generating at least one augmented query by performing augmentation on the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; the step of generating at least one response reflecting the second criterion for the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; and the step of generating at least one response reflecting the second criterion for the at least one augmented query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor.

[0022] In an embodiment, in the step of generating at least one question-response pair satisfying the first criterion, at least one question-response pair satisfying the second criterion can be generated by using at least one of the at least one user query and the at least one response generated by reflecting the second criterion for the at least one user query, and the at least one augmented query and the at least one response generated by reflecting the second criterion for the at least one augmented query.

[0023] In an embodiment, at least one question-response pair satisfying the second criterion may include at least one of at least one question-response pair configured to have at least one user query and at least one response reflecting the second criterion for the at least one user query, and at least one question-response pair configured to have at least one augmented query and at least one response reflecting the second criterion for the at least one augmented query.

[0024] In an embodiment, the step of generating at least one question-response pair satisfying the second criterion may be the step of generating at least one question-response pair reflecting the user's response preference from the at least one document and the at least one user query.

[0025] In an embodiment, at least one question-response pair satisfying the second criterion may be configured such that at least one question and at least one response reflecting the user's response preference for the at least one question form a pair.

[0026] In an embodiment, the method further includes the step of storing the at least one data set in at least one memory, and the at least one question-answer pair may include at least one of the at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion.

[0027] In an embodiment, the method further comprises the steps of specifying at least one target model for learning and training the at least one target model using the at least one data set stored in the at least one memory, wherein the at least one data set may be constructed to include at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion.

[0028] In an embodiment, in the step of training the at least one target model, fine-tuning can be performed on the at least one target model by using at least one of the at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion.

[0029] A data generation system according to the present invention, comprising a memory configured to store executable instructions and one or more processors configured to perform operations by executing one or more instructions, receives at least one user query regarding at least one document from a user terminal, processes the at least one document and the at least one user query as input to at least one artificial intelligence model, generates at least one question-response pair satisfying at least one pre-set criterion from the at least one document and the at least one user query using the at least one artificial intelligence model, and constructs at least one data set using the at least one question-response pair.

[0030] A program according to the present invention is a program that is executed by one or more processes in an electronic device and can be stored on a computer-readable recording medium, and may include instructions for performing the steps of: receiving at least one user query regarding at least one document from a user terminal; processing the at least one document and the at least one user query as input to at least one artificial intelligence model; generating at least one question-response pair satisfying at least one predetermined criterion from the at least one document and the at least one user query using the at least one artificial intelligence model; and constructing at least one data set using the at least one question-response pair. Effects of the invention

[0031] As described above, according to the data generation system and control method of the present invention, domain-specific question-response pairs and question-response pairs reflecting the user's response preferences can be automatically generated using an artificial intelligence model, and a target model for training can be fine-tuned based on the generated question-response pairs. Through this, the present invention can secure a model that is specialized for the domain and simultaneously generates user-customized responses according to the user's response preferences.

[0032] Furthermore, according to the data generation system and control method of the present invention, at least one of high-quality domain-specific data and data reflecting user preferences can be automatically generated using an artificial intelligence model, and a target model for training can be precisely tuned based on this. Through this, users can generate various data with minimal manpower, cost, and time, and receive a model built and tuned to suit their needs. In other words, users can receive a model that generates customized responses based on user preferences while simultaneously improving response accuracy and performance in the domain.

[0033] Furthermore, according to the data generation system and control method of the present invention, a user environment can be provided that automates everything from the generation of domain-specific data and data reflecting user preferences to model training. Through this, users can easily and simply generate desired data quickly and safely, thereby increasing data productivity and efficiency, and automatically build a customized, optimized model.

[0034] Furthermore, according to the data generation system and control method of the present invention, data generation results generated by reflecting user input and preferences, and training and performance evaluation results of a model fine-tuned with the generated data can be visualized and output. That is, by enhancing data understanding based on visualization in the present invention, the user is provided with an environment in which they can intuitively recognize the generated data and the training and performance evaluation results of the fine-tuned model. Brief explanation of the drawing

[0035] FIG. 1 illustrates an example of a block diagram of a computing system in which the present invention can be implemented. FIG. 2 illustrates an example of a block diagram of a computing device that may be included in a user computing device, a server computing system, and a training computing system, as an embodiment of a computing system in which the present invention can be implemented. Figure 3 illustrates an example of a block diagram from another perspective of a computing device, which is one of the components of a computing system. FIG. 4a is a conceptual diagram illustrating a data generation system according to the present invention. FIG. 4b is a block diagram illustrating a data generation system according to the present invention. FIG. 5 is a flowchart illustrating a control method for a data generation system according to the present invention. FIGS. 6, FIGS. 7, FIGS. 8, FIGS. 9, FIGS. 10, FIGS. 11, FIGS. 12, FIGS. 13, FIGS. 14a, FIGS. 14b, FIGS. 15a, FIGS. 15b, FIGS. 16, FIGS. 17, FIGS. 18, FIGS. 19 and FIGS. 20 are conceptual diagrams for explaining a control method of a data generation system according to the present invention. Specific details for implementing the invention

[0036] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components are assigned the same reference number regardless of the drawing symbols, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably solely for the ease of drafting the specification and do not inherently possess distinct meanings or roles. Furthermore, in describing embodiments disclosed in this specification, if it is determined that a detailed description of related prior art could obscure the essence of the embodiments disclosed in this specification, such detailed description will be omitted. Additionally, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification; the technical concept disclosed in this specification is not limited by the attached drawings, and it should be understood that they include all modifications, equivalents, and substitutions that fall within the spirit and technical scope of the present invention.

[0037] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. These terms are used solely for the purpose of distinguishing one component from another.

[0038] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0039] A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0040] In this application, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0041] Hereinafter, the present invention will be examined in more detail with reference to the attached drawings. FIG. 1 illustrates an example of a block diagram of a computing system in which the present invention can be implemented. FIG. 2 illustrates an example of a block diagram of a computing device that may be included in a user computing device, a server computing system, and a training computing system as an embodiment of a computing system in which the present invention can be implemented. FIG. 3 illustrates an example of a block diagram of a computing device in another aspect that is one of the components of a computing system.

[0042] Furthermore, FIG. 4a is a conceptual diagram for explaining a data generation system according to the present invention, and FIG. 4b is a block diagram for explaining a data generation system according to the present invention. FIG. 5 is a flowchart for explaining a control method of a data generation system according to the present invention, and FIG. 6, FIG. 7, FIG. 8, FIG. 9, FIG. 10, FIG. 11, FIG. 12, FIG. 13, FIG. 14a, FIG. 14b, FIG. 15a, FIG. 15b, FIG. 16, FIG. 17, FIG. 18, FIG. 19 and FIG. 20 are conceptual diagrams for explaining a control method of a data generation system according to the present invention.

[0043] Meanwhile, FIG. 1 illustrates an example of a block diagram of a computing system in which the present invention can be implemented. In this regard, the data generation system according to the present invention can be implemented through a computing device described below and can perform data processing related to the control method of the data generation system described above.

[0044] Referring to FIG. 1, a computing system (10000) that performs a method for automatically generating domain-specific data (or domain-specific question-answer pairs or a dataset including domain-specific question-answer pairs, etc.) and data reflecting user preferences (or question-answer pairs reflecting user preferences or a dataset including question-answer pairs reflecting user preferences, etc.) according to one embodiment of the present invention may include at least one computing device. In this case, the at least one computing device may be a single processor or a multi-processor computing device.

[0045] The components of at least one computing device of the present invention may include various hardware components such as one or more processors, memory, other hardware, and a system bus (not shown) that connects various system components so that they can transmit and receive data to and from each other (e.g., telecommutatively connected, physically connected, electrically connected), and the components of at least one computing device are not limited thereto and may be very diverse.

[0046] Meanwhile, at least one computing device included in a computing system (10000) that performs a method for automatically generating domain-specific data and data reflecting user preferences may be connected to communicate via a network (1070). For example, at least one computing device included in the computing system (10000) may be clustered or part of a local area network (LAN). Additionally, at least one computing device may be part of a wide area network (WAN) or connected to at least one of a client-server network and a peer-to-peer network within the cloud.

[0047] Meanwhile, when at least one computing device is used in at least one of a network environment and a cloud computing environment, the at least one computing device may be connected to at least one of a public and private network through a network interface or adapter. In one embodiment, other communication connection devices, such as a modem, may be used to establish communication through the network. The modem may be at least one of an internal modem and an external modem, and may be connected to a system bus through a network interface or a specific mechanism, etc. A wireless network component consisting of an interface and an antenna may be coupled to the network through a device such as an access point, a peer computer, etc. In the present invention, the method of connecting at least one computing device to communicate through the network (1070) is not limited, and it may be connected to communicate in a manner different from the described example.

[0048] Furthermore, other computer-type devices and / or systems not shown in FIG. 1 may also interact technically with at least one computing device or other system through one or more connections to the network (1070) via a network interface. Here, the network interface may include network interface equipment such as a physical network interface controller (NIC) or a virtual network interface (VIF).

[0049] The network (1070) of the present invention may include various forms such as the Internet, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™ RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, Wireless USB (Wireless Universal Serial Bus), etc., and in the present invention, data transmission may be performed based on standard communication protocols such as TCP / IP, HTTP, SSL, etc.

[0050] A computing system (10000) that performs a method for automatically generating domain-specific data and data reflecting user preferences according to the present invention may include at least one of a user computing device (1010), a training computing system (1050), and a server computing system (1030).

[0051] A user computing device (1010) according to the present invention may be understood as a computing device comprising at least one and / or at least one processor (1011) and memory (1012) that perform a method of automatically generating data reflecting domain-specific data and user preferences. For example, the user computing device (1010) may include at least one computing device among a smartphone, a smart TV, a laptop computer, a desktop computer, a digital broadcasting terminal, a PDA (personal digital assistants), a PMP (portable multimedia player), a navigation device, a slate PC, a tablet PC, an ultrabook, a wearable device (e.g., a smartwatch, a smart glass, and a head-mounted display).

[0052] At least one and / or at least one processor (1011) constituting the user computing device (1010) may include one or more general-purpose processors and / or one or more special-purpose processors. For example, at least one and / or at least one processor (1011) constituting the user computing device (1010) may be composed of at least one or a plurality of electrically connected processors among a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), an arithmetic logic unit (ALU), a floating-point arithmetic unit (FPU), an application integrated circuit, an application semiconductor (ASIC), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0053] Furthermore, at least one and / or at least one processor (1011) may be configured to execute computer-readable instructions contained in memory (1012) and / or other instructions described herein.

[0054] The memory (1012) constituting the user computing device (1010) according to the present invention may include volatile memory, non-volatile memory, fixed media, removable media, magnetic media, optical media, semiconductor media and / or other types of physically durable storage media.

[0055] For example, memory (1012) may include one or more non-transient / transient computer-readable storage media such as RAM, ROM, HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), EEPROM, EPROM, flash memory device, magnetic disk, etc., and may include web storage of a server that performs the storage function of memory over the internet. Such memory (1012) may store data and instructions necessary for the at least one and / or at least one processor (1011) to perform the operation of an application for automatically generating domain-specific data and data reflecting user preferences.

[0056] A user computing device (1010) may include one or more user input components (1021) that detect user input. For example, the user input component (1021) may also be referred to as a user interface module. The user input component (1021) may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not limit the type of user input component (1021). In this case, the user input component (1021) in the present invention does not necessarily mean a hardware means, but can be understood as a channel for receiving input from a user. Meanwhile, the user of the present invention may refer to an automated agent, script, playback software, etc., that operates on behalf of one or more people.

[0057] A user can interact with a computing system (10000) including at least one computing device through input text, touch, voice, movement, computer vision, gestures and / or other forms of input / output using a user input component (1021). For example, the user input component (1021) may include one or more of a command line interface (CLI), a graphical user interface (GUI), a natural user interface (NUI), a voice command interface and / or other user interface (UI) representations.

[0058] Between the user input component (1021) and the user computing device (1010), one or more application programming interface (API) calls may be made based on user input received from the user interface and / or network.

[0059] Here, the expression "based on" may be interpreted to include cases where it is based on the use of a specific configuration, modified from, derived from, influenced by, dependent on, or otherwise derived from a specific configuration. In some embodiments, an API call may be configured for a specific API, which may be interpreted or converted into an API call configured for another API. Here, an API may refer to a defined interface or connection between computers or between computer programs.

[0060] In one embodiment, the user computing device (1010) may store at least one machine learning model (1020). For example, the user computing device (1010) may be various machine learning models, such as a plurality of neural networks (e.g., deep neural networks) that perform automatic generation of domain-specific data and data reflecting user preferences using user input and user preferences (or user response preferences), or other types of machine learning models including non-linear models and / or linear models, and may be composed of a combination thereof.

[0061] According to an embodiment of the present invention, a user computing device (1010) may perform a method of automatically generating domain-specific data and data reflecting user preferences using a local or / and external machine learning model (1020). Alternatively, the user computing device (1010) may perform a method of automatically generating domain-specific data and data reflecting user preferences using a machine learning model (1040) provided by a server.

[0062] In addition, according to another embodiment of the present invention, a server computing system (1030) communicating with a user computing device (1010) can provide domain-specific data and data reflecting user preferences to the user computing device (1010) on an application or / and the web in accordance with a user's request received through the user computing device (1010).

[0063] In addition, according to another embodiment of the present invention, by performing a method to automatically generate domain-specific data and data reflecting user preferences by linking at least a part of a user computing device (1010) and a server computing system (1030) with each other, data reflecting domain-specific data and user preferences can be provided to the user.

[0064] Additionally, according to various embodiments of the present invention, a user computing device (1010) and / or a server computing system (1030) can learn a machine learning model (1020, 1040) performed in a method of automatically generating domain-specific data and data reflecting user preferences through interaction with a training computing system (1050) that is communicatedly connected via a network (1070). In this case, the training computing system (1050) may be a computing system separate from the server computing system (1030). Alternatively, in some embodiments, the training computing system (1050) may be part of the server computing system (1030) or part of the user computing device (1010).

[0065] Meanwhile, the server computing system (1030) may include at least one processor (1031) and memory (1032). Here, the processor (1031) may be composed of at least one or a plurality of electrically connected processors among a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), an application integrated circuit, an application semiconductor (ASIC), an arithmetic logic unit (ALU), a floating-point arithmetic unit (FPU), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions. For example, at least one processor (1031) may include a circuit and a transistor configured to execute instructions from memory (1032).

[0066] The memory (1032) constituting the server computing system (1030) according to the present invention may include volatile memory, non-volatile memory, fixed media, removable media, magnetic media, optical media, semiconductor media, and / or other types of physically durable storage media. For example, the memory (1032) may include one or more non-transient / transient computer-readable storage media such as RAM, ROM, HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), EEPROM, EPROM, flash memory device, magnetic disk, etc., and combinations thereof, and may include web storage of a server that performs the storage function of memory over the internet. Additionally, the server computing system (1030) may further include a data storage (data store). For example, the data storage may be composed of at least one of a relational database, a NoSQL database, a data warehouse, and a local file system.

[0067] In the memory (1032) constituting the server computing system (1030) according to the present invention, data and instructions necessary for the at least one processor (1031) to perform the operation of an application for automatically generating domain-specific data and data reflecting user preferences may be stored.

[0068] In one embodiment, the server computing system (1030) may be composed of a single device or a plurality of computing devices, and may be configured to operate according to a sequential or parallel computing architecture. Additionally, a distributed processing system may be configured with a plurality of networked devices.

[0069] Meanwhile, the training computing system (1050) may include at least one processor (1051) and memory (1052). The model trainer (1060) is a logical component that executes the training of at least one machine learning model (1020, 1040) and may be implemented in the form of hardware, firmware, or software. For example, the model trainer (1060) may be executed by the processor (1051) after loading training data (1061) stored in a storage device into memory (1052). For example, the model trainer (1060) may be configured to execute one or more operations (e.g., model training, model reconstruction, model validation, model testing) on ​​at least one machine learning model.

[0070] The machine learning model of the present invention may include at least one of a statistical model, an algorithm, a neural network (NN), a convolutional neural network (CNN), a generative neural network (GNN), a Word2Vec model, a Bag of Words model, a TF-IDF (document frequency-inverse document frequency) model, a GPT (Generative Pre-trained Transformer) model (or other autoregressive models), a PPO (Proximal Policy Optimization) model, a nearest neighbor model (e.g., a k-nearest neighbor model), a linear regression model, a K-means clustering model, a Q-learning model, a TD (Temporal Difference) model, a Deep Adversarial Network model, and all other types of models further described herein.

[0071] Specifically, the model trainer (1060) may execute operations to train a machine learning model, and said operations may include at least one of adding, removing, and modifying model parameters. At this time, the training of the machine learning model may be at least one of supervised learning, semi-supervised learning, and unsupervised learning. In one embodiment, the training of the machine learning model may include the step of repeatedly inputting training data (1061) based on epochs and repeatedly performing the machine learning model training process configured in this way. Here, an epoch may refer to a unit in which the entire set of training data (1061) undergoes forward and backpropagation processing once. In some implementations, different levels of training methods (e.g., supervised learning, semi-supervised learning, unsupervised learning) may be used for different epochs.

[0072] The training data (1061) of the present invention may include input data and / or data previously output from at least one machine learning model (e.g., recursive learning feedback).

[0073] At least one parameter of a machine learning model may include at least one of a seed value, a model node, a model layer, an algorithm, a function, connections between different machine learning models, connections between parameters, machine learning model constraints, and other digital components that influence the output of the machine learning model. In this case, model connections between different machine learning models may include or represent relationships between model parameters and / or models, which may be dependent or interdependent, hierarchical, and / or static or dynamic. The combinations and configurations of model parameters described herein may be too complex to be maintained or utilized by human cognitive abilities.

[0074] In the present invention, the machine learning parameters described according to the embodiments are not limited, and a single machine learning model may further include a plurality of model parameters.

[0075] Meanwhile, FIG. 2 illustrates an example of a block diagram of a computing device (1100) that may be included in a user computing device (1010), a server computing system (1030), and a training computing system (1050), as an embodiment of a computing system (10000) in which the present invention can be implemented.

[0076] As illustrated in FIG. 2, the computing device (1100) may include at least one application (e.g., Application 1 to Application N), and each of the at least one application may include a machine learning library and a model execution environment for performing a method to automatically generate machine learning-based domain-specific data and data reflecting user preferences. The at least one application included in the computing device (1100) may communicate with sensors, context managers, device state managers, or additional component(s) within the computing device (1100) via an Application Programming Interface (API). In one embodiment, the at least one application may interface with device components, such as receiving sensor data or state data or transmitting prediction results to an output device via a public or private API.

[0077] Meanwhile, FIG. 3 illustrates an example of a block diagram in another aspect of a computing device (1200), which is one of the components of a computing system (10000) that performs a method for automatically generating domain-specific data and data reflecting user preferences according to an embodiment of the present invention.

[0078] A computing device (1200) according to the present invention may include at least one application (e.g., Application 1 to Application N), and at least one application may communicate with a central intelligence layer (1210). Each application may interact with a shared model within the central intelligence layer (1210) through an API (e.g., a common API).

[0079] The central intelligence layer (1210) includes one or more machine learning models and may share them among multiple applications or provide them independently to each. In one embodiment, the central intelligence layer (1210) may be integrated as part of an operating system or implemented as a separate logical layer.

[0080] Additionally, the central intelligence layer (1210) can communicate with the central device data layer (1220). The central device data layer (1220) can store at least one question-answer pair included in each of the domain-specific data and user preference-reflecting data stored within the computing device (1200), and provide this as input data required for training (e.g., fine-tuning) at least one target model. Each device component (e.g., sensor, state manager, etc.) can communicate with the central device data layer (1220) via a private API, etc.

[0081] The technology described herein may be composed of a single or multiple computing devices, and a machine learning model that performs a method for automatically generating domain-specific data and data reflecting user preferences may be executed sequentially or in parallel on a single component or multiple distributed components. Data storage, machine learning models, and applications may be distributed and operated locally or over a network, and these configurations can be flexibly applied to various system architectures.

[0082] Meanwhile, the present invention relates to a data generation system capable of automatically generating high-quality data based on artificial intelligence and a method for controlling the same. The data generation system according to the present invention performs data generation based on at least one artificial intelligence model and may also be referred to as an “artificial intelligence model-based data generation platform,” an “artificial intelligence model-based data generation service,” an “artificial intelligence model-based data foundry platform,” or an “artificial intelligence model-based data foundry service.” For example, the artificial intelligence model utilized (or used, etc.) to automatically generate high-quality data in the present invention may include at least one of a Large Language Model (LLM), a Super-Large Foundation Model, and a Generative AI Model. The data generation system according to the present invention may be a system that automatically generates domain-specific data and data reflecting user preferences based on (or reflecting) user input (e.g., documents and queries, etc.) and user preferences (e.g., response preference, response style, scope of information usage, etc.). In addition, the data generation system according to the present invention may be a system that automatically performs everything from the automatic generation of domain-specific data and data reflecting user preferences to fine-tuning of the target model for training.

[0083] The user (or customer, company, client company, etc.) described above may possess a user account already registered in the data generation system (or data generation platform) according to the present invention. In this case, the account may be created through at least one page (or screen) linked to the data generation system. Alternatively, the account may also be created in at least one other system linked to the data generation system. However, in this specification, the system in which the user account was issued is not distinguished separately, and all accounts capable of using various services (or functions) provided by the data generation system (100) according to the present invention are referred to as “accounts already registered in the data generation system.”

[0084] Meanwhile, the data generation system according to the present invention includes at least one artificial intelligence model, and the present invention aims to provide a platform that automates everything from the generation of high-quality data specialized for various domains and / or tasks to model training.

[0085] In this case, at least one artificial intelligence model mentioned in the present invention may include various models that can be utilized depending on various situations or purposes. For example, the artificial intelligence model of the present invention may include at least one of a machine learning (ML) model, a deep learning model, a deep neural network (DNN), a language model (LM), a large language model (LLM), a super-large foundation model, a generative AI model, a transformer-based model, a supervised learning (SL) model, a reinforcement learning (RL) model, and a special purpose model (e.g., a time series forecasting model (e.g., ARIMA model, SARIMA model, etc.), a time series foundation model, a graph neural network (GNN), a multimodal model, a natural language processing (NLP) model, a computer vision model, a speech recognition / synthesis model, a recommendation system model, etc.).

[0086] Furthermore, the artificial intelligence model used in the data generation system according to the present invention may be implemented with a single or multiple components. More specifically, the artificial intelligence model used in the present invention may be implemented with a single component (1) or with multiple components (2, 3, 4 or more, etc.).

[0087] In this regard, when multiple artificial intelligence models are implemented in the present invention, the multiple artificial intelligence models may be implemented as models of the same type (e.g., each implemented as a first large-scale language model, a second large-scale language model, etc.). In this case, the multiple artificial intelligence models may be implemented to perform different functions (or roles) in the present invention (e.g., the first large-scale language model performs a first function, and the second large-scale language model performs a second function).

[0088] Alternatively, multiple AI models may each be implemented as different models (for example, at least one AI model is implemented as a machine learning model, and at least one other AI model is implemented as a large-scale language model). In this case, the multiple AI models may be physically separated models (or modules) or multiple functional units within a single piece of hardware. However, this specification does not limit the implementation of AI models to a singular or multiple forms, and for convenience, they will be described as "at least one AI model."

[0089] Meanwhile, as illustrated in FIG. 4a, the data generation system (100) according to the present invention may include at least one of an input unit (110), an output unit (120), a communication unit (130), a storage unit (140), a data processing unit (150), a data generation unit (160), an evaluation unit (170), and a control unit (180). However, the components of the data generation system (100) according to the present invention are not limited thereto and may further include various hardware components that perform the same or similar roles as described in the description of the present specification.

[0090] Although not illustrated, the data generation system (100) according to the present invention may include one or more processors, and such processors may include one or more general-purpose processors and / or one or more special-purpose processors (e.g., digital signal processor, tensor processing unit (TPU), graphics processing unit (GPU), neural network processing unit (NPU), application integrated circuit, application semiconductor (ASIC), field programmable gate array (FPGA), quantum processing unit (or quantum processor, QPU), etc.). One or more processors may be configured to execute instructions, computer-readable instructions, and / or other instructions described herein that are stored (or included) in the storage unit (140). The data generation system and the control method according to the present invention may perform the data processing described below in cooperation with memory and at least one processor. The processor may perform a series of operations and data processing using data and information stored in memory. In this case, memory may be a component of the storage unit (140).

[0091] In addition, the data generation system (100) according to the present invention can perform data processing and computation processes using quantum gates, quantum entanglement, and quantum superposition states, taking into consideration implementation in a quantum computer environment. For example, the present invention can perform parallel computations based on qubits, and such quantum computations can operate complementarily with existing classical computers.

[0092] Such quantum computers may include parallel computation using qubits and high-speed data processing devices utilizing quantum entanglement, and hardware-based computational optimization using FPGAs and ASICs is possible. In addition, quantum computers may utilize quantum processors capable of qubit-based parallel computation, and data processing efficiency can be improved through a hybrid structure with existing classical computers.

[0093] Meanwhile, the input unit (110) can be configured in various ways as a means of data input. For example, the input unit (110) can be configured to receive user input. The input unit (110) can be configured to receive user input from a user terminal (10). Here, “receiving input” may mean receiving an input signal (or selection signal) corresponding to the user’s input based on input made by the user through the input unit configuration provided in the user terminal (10).

[0094] Here, the user terminal (10) may include at least one of a mobile phone, a smartphone, a notebook computer, a laptop computer, a slate PC, a tablet PC, an ultrabook, a desktop computer, a digital broadcasting terminal, a PDA (personal digital assistants), a PMP (portable multimedia player), a navigation device, and a wearable device (e.g., a smartwatch, a smart glass, a head-mounted display).

[0095] In addition, the input unit (110) in the present invention does not necessarily mean a hardware means, but can be understood as a channel for receiving input from a user.

[0096] The input unit (110) may also be referred to as a user interface module. The input unit (110) may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not limit the type of input unit (110).

[0097] Here, user input may include documents, text, images (or videos), voice, etc. In this case, the data generation system (100) may further include a module that converts voice into text.

[0098] Next, the output unit (120) can output information through an output unit configuration (e.g., a display unit, a touch screen, a speaker, etc.) provided in a user terminal (10) linked to the data generation system (100) according to the present invention. For example, the output unit (120) can output at least one page (1000, or service page) linked to the data generation system (100) according to the present invention to the display unit of the user terminal (10). Additionally, the output unit (120) does not necessarily mean a hardware means, but can be understood as a channel for outputting results to a user.

[0099] Next, the communication unit (130) may be connected via a wireless or wired network to a user terminal (10), a server (e.g., a central server, an external server, etc.), a device, and at least one network, etc., to receive or transmit overall data and information necessary for the operation of the data generation system (100) according to the present invention.

[0100] The communication unit (130) can support various communication methods depending on the communication standard of the communicating device.

[0101] For example, the communication unit (130) may be configured to communicate with a communication target using at least one of the following technologies: WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™ RFID (Radio Frequency Identification), Infrared Communication (Infrared Data Association; IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, and Wireless USB (Wireless Universal Serial Bus).

[0102] Next, the storage unit (140, or memory) serves to store various data related to the present invention and may include one or more non-transient computer-readable storage media that can be read and / or accessed by at least one of one or more processors.

[0103] One or more computer-readable storage media may include volatile and / or non-volatile storage components, such as optical, magnetic, organic, or other memory or disk storage devices. In some examples, the storage unit (140) may be implemented using a single physical device (e.g., one optical, magnetic, organic, or other memory or disk storage device), whereas in other examples, the storage unit (140) may be implemented using two or more physical devices.

[0104] The storage unit (140) may include computer-readable instructions and additional data. The storage unit (140) may include a storage necessary to perform at least some of the methods, scenarios, and techniques described herein and / or at least some of the functions of the device and network.

[0105] Furthermore, at least a portion of the storage unit (140) may be a cloud storage or a cloud server. At least a portion of the data corresponding to user input and training data received from the input unit (110) may be stored in the storage unit (140).

[0106] Additionally, the storage unit (140) may store at least one data collected (or received) from various sources (e.g., a web corpus, a document corpus, a database (DB) website, an API, a server linked to the data generation system (100), a central server, an external server, cloud storage, a user terminal (10), a large dataset, etc.). For example, the storage unit (140) may store at least one and / or at least one or more (i.e., multiple) documents collected from at least one of the various sources (e.g., a user terminal (10)). In this case, the collected documents may include documents related to at least one and / or at least one domain (or field).

[0107] That is, the storage unit (140) is sufficient as a space where information necessary for the operation of the data generation system (100) according to the present invention is stored, and it can be understood that there are no restrictions on the physical space.

[0108] Furthermore, the storage unit (140) may store a computer program including computer program instructions. Furthermore, the storage unit (140) may store a computer program including computer program instructions that control the operation of the system (100) or control the operation of the control unit (180) when loaded into the processor of the system (100).

[0109] Next, the data processing unit (150) may be configured to perform data preprocessing and / or data cleansing on the data required to automatically generate domain-specific data and data reflecting user preferences.

[0110] The data processing unit (150) may be configured to perform preprocessing and / or refinement on at least one document collected (or received) from various sources (or various data sources).

[0111] To this end, the data processing unit (150) may include a document processing module (or model, engine, etc.) or an algorithm that analyzes (or understands, interprets, etc.) data (e.g., documents) and converts it into a structured form. For example, the document processing module may structure paragraphs (or sections), tables (or tables), sections, etc. included in the document, assign meta-information (or data) per document, or perform PII (Personal Identifiable Information) data masking. In the present invention, the document processing module may also be named a “document processing model,” a “document processing engine,” a “document parser module,” a “document parser model,” a “document parser engine,” a “document parsing module,” a “document parsing model,” or a “document parsing engine.”

[0112] Additionally, the data processing unit (150) may include an OCR (Optical Character Recognition) model (or OCR module, OCR engine, etc.) that recognizes text (e.g., characters, symbols, etc.) contained in a document and / or image through automated data extraction and converts it into machine-readable text data.

[0113] Additionally, the data processing unit (150) may include a document search module (or document search model, document search engine, etc.) configured to search for documents related to (or associated with) at least one query from various sources (e.g., a web corpus, a document corpus, a database (DB) website, an API, a server linked to the data generation system (100), a central server, an external server, cloud storage, a user terminal (10), a large dataset, etc.).

[0114] Next, the data generation unit (160) may be configured to perform the role of generating at least one data set (400) including at least one question-response pair in response to a user request (or input) received through a user terminal (10, or at least one page (1000) associated with the data generation platform).

[0115] To this end, the data generation unit (160) may include at least one artificial intelligence model (e.g., a large-scale language model (LLM)) configured to perform the role of generating data based on user input. Alternatively, the data generation unit (160) itself may correspond to at least one artificial intelligence model. However, for convenience of explanation, this specification describes them without distinguishing between them.

[0116] In one embodiment, the data generation unit (160) may generate at least one data set (400) including at least one of domain-specific data (or domain-specific question-answer pairs or a data set including domain-specific question-answer pairs, etc.) and data reflecting user preferences (or question-answer pairs reflecting user preferences or a data set including question-answer pairs reflecting user preferences, etc.) based on at least one query (or user query) for at least one document. More specific details regarding data generation will be described later.

[0117] Next, the evaluation unit (170) may be configured to perform the role of evaluating at least one data set (400) generated through the data generation unit (160). Additionally, the evaluation unit (170) may be configured to perform the role of evaluating the performance (or capability) of at least one model learned using the data set (400).

[0118] To this end, the evaluation unit (170) may include at least one artificial intelligence model configured to perform automatic evaluation (e.g., evaluation of accuracy, logic, style matching, etc.) on question-answer pairs included in at least one data set (400). Additionally, the evaluation unit (170) may include at least one artificial intelligence model configured to perform an evaluation of the performance of at least one model learned using at least one data set (400). In one embodiment, at least one artificial intelligence model included in the evaluation unit (170) may be a large-scale language model (LLM) based model. Such an artificial intelligence model included in the evaluation unit (170) may also be referred to as an “evaluator model.” More specific details regarding this will be described later.

[0119] Next, the control unit (180) can perform the role of controlling the overall operation of the data generation system (100) related to the present invention. The control unit (180) can process signals, data, information, etc. that are input or output through the components of the data generation system (100) described above, or perform a series of data processing to provide or process appropriate information and functions to the user. The control unit (180) can be physically implemented by the processor described above.

[0120] Meanwhile, although not illustrated, the data generation system (100) according to the present invention may include at least one and / or at least one specialized model (SM).

[0121] In one embodiment, the expert model (SM) may refer to an artificial intelligence model that has undergone optimized learning for a specific purpose. Additionally, the expert model (SM) may refer to an artificial intelligence model that has undergone optimized learning for a specific purpose. That is, the expert model (SM) may be an artificial intelligence model trained according to specialized training data and methods to achieve a predetermined purpose.

[0122] Such specialized models (SM) may include a learned specific sLLM (including MoELM and / or DMoE models), a general MoE model (NM) and / or an external model (EM)), etc. Additionally, the specialized model may include a specialized module model according to an embodiment of the present invention disclosed in the control method of the data generation system according to the present invention described below. The sLLM, general MoE model (NM), external model (EM) and / or specialized model (SM) as described above may be a secondary model (S) capable of performing a specific task according to the control and management of the master model (P) of the AI ​​agent model (i.e., an orchestrator (OCT) and / or router (RT), etc.). Here, the router (RT) may be an artificial intelligence module that selectively determines a model (hereinafter referred to as a domain-specific model) that executes data processing operations optimized for a specific domain.

[0123] In this regard, the data generation system according to the present invention may include a Mixture of Experts (MoE) structure in which a plurality of expert models are selectively activated according to modality characteristics. Alternatively, at least one artificial intelligence model according to the present invention may include a MoE structure in which a plurality of expert models (or expert modules, specialized models, expert processing models, etc.) and / or sub-models are dynamically activated according to query characteristics and document context. Such a MoE structure enables the simultaneous assurance of accuracy and universality in input interpretation and output generation (e.g., responses, question-answer pairs, etc.) by selectively deploying the most suitable expert model for various types of inputs (e.g., queries and documents, etc.) or domain situations.

[0124] Here, MoE may refer to the architecture of a machine learning model that solves complex problems by combining multiple expert models. Such a MoE may include expert models, which are multiple small networks designed to learn different parts and / or different characteristics of given data and perform data processing operations accordingly, and a gating network that evaluates the performance of each expert model and, based on this, determines which expert model is most suitable for assigning a specific task based on given data.

[0125] Thus, according to the MoE architecture, a gating network that acquires predetermined input data determines probabilistic or deterministic task assignments for each expert model, and the selected expert models perform their respective tasks and return the results, thereby enabling data processing for a specific task.

[0126] Meanwhile, FIG. 4b is a block diagram illustrating the overall flow of a data generation system (100, or data generation platform) according to the present invention.

[0127] Referring to FIG. 4b, the control unit (180) may receive user input necessary to generate data from the user terminal (10) (e.g., at least one document (401), at least one query (402), etc.). In the present invention, receiving user input from the user terminal (10) may be interpreted as having the meaning of “receiving user input (or selection) through at least one page output to the user terminal (10)” or “receiving user input (or selection) from the user terminal (10) on which at least one page has been output.”

[0128] For example, the control unit (180) may receive at least one document (401) related to at least one domain and at least one query (or user query, 402) regarding at least one document (401) based on (or based on) user input received from the user terminal (10). In this case, the control unit (180) may identify a specific domain from the document (401) based on the fact that the document (401) related to at least one domain has been received.

[0129] Alternatively, at least one document (401) may include at least one document retrieved through a document search module included in the data processing unit (150). For example, let us assume that at least one query (402) is received from a user terminal (10). The data generation system (100) according to the present invention may use a document search module to search (or identify) at least one document (401) related (or associated) with at least one query (402) from various sources, and generate at least one response to at least one query (402) based on the retrieved at least one document (401). In this case, at least one query (402) may be a query related to at least one domain. A data generation system (100) according to the present invention can identify at least one domain from the query (402) based on the receipt of the query (402) associated with at least one domain, and search for at least one document (401) associated with the query (402) and the domain through a document search module.

[0130] In addition, the “at least one document” mentioned in the present invention may be implemented as a singular (1) or as a plural (e.g., 2, 3, 4 or more). Accordingly, in this specification, the “at least one document” may also be expressed as “document,” “plural documents,” “document related to a domain,” “plural documents related to a domain,” “plural documents related to each of a plurality of domains,” “document related to a query,” “plural documents related to a query,” “plural documents related to a plurality of queries,” or “plural documents related to each of a plurality of queries.”

[0131] In addition, the “at least one query” mentioned in the present invention may be implemented as a singular (1) or as a plural (e.g., 2, 3, 4 or more). Accordingly, in this specification, the “at least one query” may also be expressed as “query,” “plural query,” “query related to a domain,” “plural query related to a domain,” “plural query related to each of a plurality of domains,” “query for a document,” “plural query for a document,” “plural query for a plurality of documents,” or “plural query for each of a plurality of documents.”

[0132] In addition, the “at least one response” mentioned in the present invention may be implemented as a singular (1) or as a plural (e.g., 2, 3, 4 or more). Accordingly, in this specification, the “at least one response” may also be expressed as “response,” “multiple responses,” “response to a query,” “multiple responses to a query,” “multiple responses to multiple queries,” or “multiple responses to each of multiple queries.”

[0133] In addition, the control unit (180) can input a document (401) and a query (402) regarding the document (401) into the data generation unit (160). Here, inputting into the data generation unit (160) can also be interpreted to mean “inputting into at least one artificial intelligence model that performs the role of generating data.” Furthermore, in addition to the document (401) and the query (402), the input to this artificial intelligence model may further include user response preference information to generate a customized response based on user preference. This user response preference may be information stored in a designated storage (e.g., a storage unit (140) or at least one memory, etc.) linked to a user account using the data generation platform. Alternatively, it may be information determined based on user input received from the user terminal (10). More specific details regarding the response preference will be described later.

[0134] The data generation unit (160) can generate at least one question-response pair that satisfies at least one previously set criterion.

[0135] In one embodiment, a data generation unit (i.e., an artificial intelligence model, 160) can generate at least one question-response pair (411) satisfying a first criterion among at least one pre-set criterion (or a plurality of pre-set criters) from the input document (401) and the query (402) regarding the document (401). In the present invention, generating a question-response pair satisfying the first criterion can also be interpreted to mean “generating a question-response pair specialized in at least one domain.” Such a question-response pair (411) satisfying the first criterion may be configured such that at least one query (411a) specialized in at least one domain and at least one response (411b) to the query (411a) form a pair. At this time, at least one query (411a) included in the question-answer pair (411) satisfying the first criterion may correspond to at least one of the query (402) corresponding to the user's input or the query augmented from said query (402). Additionally, at least one response (411b) to said query (411a) may correspond to at least one of the response to the query (402) corresponding to the user's input or the response to the augmented query.

[0136] In another embodiment, the data generation unit (160) can generate at least one question-response pair (412) satisfying a second criterion among a plurality of pre-set criteria from the input document (401) and the query (402) regarding the document (401). In the present invention, generating a question-response pair satisfying the second criterion can also be interpreted to mean “generating a question-response pair that reflects the user’s response preference.” Such a question-response pair (412) satisfying the second criterion may be configured such that at least one query (412a) and at least one response (412b) that reflects the user’s response preference for the at least one query (412a) form a pair. At this time, at least one query (412a) included in the question-answer pair (412) satisfying the second criterion may correspond to at least one of the query (402) corresponding to the user's input or the query augmented from said query (402). Additionally, at least one response (412b) generated by reflecting the second criterion for said query (412a) may correspond to at least one of the response to the query (402) corresponding to the user's input or the response to the augmented query.

[0137] Subsequently, the data generation unit (160, or control unit (180)) can construct (or generate, configure, etc.) at least one data set (or training data set) using at least one question-answer pair (411) satisfying a first criterion and at least one question-answer pair (412) satisfying a second criterion.

[0138] In one embodiment, the data generation unit (160) can construct a first data set (or a first training data set, 410) using at least one question-answer pair (411) that satisfies a first criterion. This first data set (410) may also be understood as a domain-specific data set containing question-answer pairs specialized in at least one domain.

[0139] In another embodiment, the data generation unit (160) may construct a second data set (or a second training data set, 420) using at least one question-answer pair (412) that satisfies the second criterion. This second data set (420) may also be understood as a user preference-reflecting data set containing question-answer pairs that reflect the user's response preferences.

[0140] However, although the above embodiment describes the configuration in which question-answer pairs satisfying the first criterion and question-answer pairs satisfying the second criterion are each included in separate data sets, it is not necessarily limited thereto. For example, it is obvious that in the present invention, question-answer pairs satisfying the first criterion and question-answer pairs satisfying the second criterion may be included in a single data set.

[0141] Furthermore, the control unit (180) can train at least one target model (500) using at least one generated data set (first data set (410), second data set (420)). The control unit (180) can perform fine-tuning on the target model (500) using at least one of at least one question-answer pair (411) satisfying a first criterion included in the first data set (410) and at least one question-answer pair (412) satisfying a second criterion included in the second data set (420). Alternatively, the control unit (180) can perform fine-tuning on the target model (500) using at least one question-answer pair (411) satisfying a first criterion included in the first data set (410) and at least one question-answer pair (412) satisfying a second criterion included in the second data set (420), respectively.

[0142] Subsequently, when training on the target model (500) is completed, the control unit (180) can obtain a trained model (510, or a fine-tuned model) through at least one data set (a first data set (410), a second data set (420)). The trained model (510) may be specialized for at least one domain or configured to generate a response that reflects the user's response preference.

[0143] Alternatively, the trained model (510) may be configured to generate a response that satisfies the first criterion and simultaneously satisfies the second criterion for a query related to at least one domain. This may mean that the trained model (510) is specialized for at least one domain and simultaneously generates a response that reflects the user's response preference for a query related to at least one domain.

[0144] Finally, the control unit (180) can provide the learned model (510) through at least one page associated with the user terminal (10) and / or the data generation platform and / or at least one user account already registered on the data generation platform.

[0145] Meanwhile, the present invention aims to solve the problem of data scarcity and to provide a data generation system and a control method thereof capable of automatically generating domain-specific data and data reflecting user preferences. More specifically, the present invention aims to provide a data generation system and a control method thereof capable of automating the process from high-quality data generation to model training. Below, we will examine the control method of the data generation system according to the present invention in more detail.

[0146] Meanwhile, in the present invention, a process of receiving at least one user query for at least one document from a user terminal may be performed (S510, see FIG. 5).

[0147] In the present invention, the order (or process) in which at least one user query is received may vary.

[0148] In one embodiment, the present invention may receive a query regarding the received document after the document has been received first. The control unit (180) may receive at least one document based on (or derived from) user input received from the user terminal (10). And, the control unit (180) may receive at least one user query regarding the received at least one document.

[0149] In another embodiment, the present invention may identify and / or receive (or collect) documents related to the received query after a query is received first. When the control unit (180) receives at least one query from the user terminal (10), it may use a document search module to search for at least one document related (or associated with) at least one query from various sources. Then, the control unit (180) may collect at least one searched document as a document related to the query.

[0150] In other words, the order in which user queries are received in the present invention may vary, and the present invention is not limited to any one such order. However, for the sake of convenience of explanation, the following description will be based on the premise that a user query regarding the received document is received after the document has been received first.

[0151] The control unit (180) may provide at least one page containing various functions for automatically generating data to the user terminal (10) in response to a user request received to automatically generate data. For example, as illustrated in FIGS. 4a and FIGS. 6, the control unit (180) may output (or provide) to the user terminal (10) at least one page (600) configured to receive (or collect) the user's document based on the selection of a graphic object (1001) linked (or linked) with a data generation request function included in at least one page (1000) linked to the data generation system (100, or data generation platform) from the user terminal (10).

[0152] And, the control unit (180) can receive at least one document related to at least one domain from the user terminal (10) on which at least one page (600) has been output. Here, a domain refers to an area where specialized knowledge, data, terms, documents, etc. belonging to (or used in) a specific industry, technology, or business field are concentrated. For example, it may include various fields such as manufacturing processes, defect analysis, medical, financial, legal, and administrative fields. That is, a domain may refer to an area (or specialized field) where knowledge, data, terms, documents, etc. belonging to a specific specialized field are concentrated.

[0153] For example, as illustrated in FIG. 6, the page (600) may be configured to receive a document related to at least one domain (e.g., public administration and social welfare services) from a user terminal (10). More specifically, the control unit (180) may output documents (601, 602, 603, 604, 605, 606) related to at least one domain stored in the data generation system (100) onto the page (600), and receive a selection from the user terminal (10) of at least some of the documents (604, 605, 606) output onto the page (600). Subsequently, the control unit (180) can receive a plurality of documents (604, 605, 606) associated with at least one domain based on the selection of a graphic object (600a) associated with a document receiving function included (or output, display, provide, etc.) in the page (600).

[0154] In this case, the documents related to at least one domain stored in the data generation system (100) may include at least one document entered by a user into the data generation system (100) and / or at least one document collected from various sources by the data generation system (100).

[0155] Alternatively, the page (600) may be configured to receive (or input) a document related to at least one domain from the user terminal (10). For example, the control unit (180) may receive at least one document related to a specific domain based on user input regarding a document related to a specific domain received through the page (600).

[0156] Subsequently, the control unit (180) may receive at least one query (or question) regarding at least one document. For example, as illustrated in FIG. 7, the control unit (180) may output at least one page (700) configured to receive (or receive) a user query regarding a document on the user terminal (10). This page (700) may include at least one document received from the user through the process S510 (e.g., a first document (604), a second document (605), a third document (606), etc.) and at least one of a query receiving area (or query input area, 710) for receiving a user query regarding said document.

[0157] The control unit (180) can receive at least one user query for at least one document based on user input received through a query receiving area (710) included in at least one page (700). More specifically, the control unit (180) can receive multiple user queries (604a, 604b, 604c) for at least one document (e.g., the first document (604)) among multiple documents (604, 605, 606) related to at least one domain based on user input received through the query receiving area (710). For example, a user can select at least one document (604) among multiple documents (604, 605, 606) included in the page (700) and input user queries (604a, 604b, 604c) for the selected document (604) through the query receiving area (710). Subsequently, the control unit (180) can receive user queries for each of the multiple documents (604, 605, 606) based on the selection of a graphic object (700a) associated with a user query receiving function included in the page (700).

[0158] Meanwhile, when a document related to at least one domain and a user query regarding said document are received, the control unit (180) may conduct a survey related to (or regarding the response preference) the user's response preference in order to specify (or judge, identify, determine, etc.) the user's response preference.

[0159] For example, as illustrated in FIGS. 8 to 10, the control unit (180) may output at least one page (800) configured to include various surveys to determine the user's response preference on the user terminal (10). The control unit (180) may determine the user's response preference based on user input received through the at least one page (800) in order to generate at least one response according to the user's response preference for at least one question.

[0160] At this time, in the present invention, at least one survey for specifying the user's response preference may include at least one of at least one survey related to the response style and at least one survey related to the scope of information usage.

[0161] For example, the response style may include at least one of i) a first response style (or list type (e.g., a format that lists key content using numbers, symbols, or bullet points to convey information quickly)), ii) a second response style (or narrative type (e.g., a format that explains content through logical description following the flow of sentences, written as complete sentences connecting background, reasons, processes, and conclusions, and is explanation-centered to aid the reader's understanding)), and iii) a third response style (or hierarchical type (e.g., a format that organizes information in a stepwise and / or tree structure from a higher concept to a lower item, where each step has a dependent relationship and systematically organizes complex content through a systematic structure)). In addition, the scope of information usage may include at least one of i) a first scope of information usage (or knowledge expansion type (e.g., a method of explaining by extensively utilizing related background knowledge, principles, similar cases, standards, etc., beyond a direct answer to a question)), ii) a second scope of information usage (or detailed information type (e.g., a method of specifically explaining details (numerical values, procedures, conditions, etc.) based on information within a given document or data)), and iii) a third scope of information usage (or key extraction type (e.g., a method of concisely summarizing by selecting only the key gist and / or key facts and / or conclusions from extensive information)).

[0162] In this regard, the control unit (180) can generate at least one candidate question-response pair using the data generation unit (160) to provide a survey for specifying the user's response preference.

[0163] Specifically, the data generation unit (160) can generate at least one candidate question-answer pair from at least one document received through a user terminal (10, or at least one page) and at least one user query regarding said at least one document. In the present invention, the statement that the data generation unit (160) generates at least one candidate question-answer pair can also be interpreted to mean that at least one artificial intelligence model (e.g., a large-scale language model (LLM)) generates at least one candidate question-answer pair.

[0164] And, the control unit (180) may provide (or output, display, etc.) at least one survey on the user terminal (10) to specify the user's response preference. More specifically, the control unit (180) may provide at least one survey including at least one candidate question-answer pair through at least one page (800).

[0165] In this case, at least some of the candidate question-answer pairs (or candidate question-answer pairs, multiple candidate question-answer pairs, etc.) generated through the data generation unit (160) may include at least one candidate question-answer pair configured to have at least one question and multiple responses with different response styles for said at least one question. Additionally, at least some of the candidate question-answer pairs may include at least one candidate question-answer pair configured to have at least one question and multiple responses with different information usage ranges for said at least one question.

[0166] First, the control unit (180) may provide at least one survey related to a response style (or at least one survey regarding a response style) on the user terminal (10) to determine (or identify, specify, determine, etc.) the response style preferred by the user. For example, as illustrated in FIGS. 8 and 9, the control unit (180) may provide at least one survey related to a response style through at least one page (800) based on the selection of a graphic object (e.g., “Start Survey”, 800a) associated with a response preference survey start function included in at least one page (800) from the user terminal (10). In this case, the survey related to a response style may include a candidate question-answer pair configured to have a question (e.g., “Please explain how the right to receive a divided pension is affected when the right to receive an old-age pension is waived.”, 900) and a plurality of responses (901, 902) with different response styles to the question (900). Alternatively, it may include a questionnaire configured to have a question (900) and a plurality of responses (or a graphic object corresponding to each of the plurality of responses (901, 902)) that have different response styles to the question (900). In one embodiment, among the plurality of responses (901, 902), the response style of the first response (901) (or includes) may correspond to a “third response style (hierarchical)”, and the response style of the second response (902) may correspond to a “first response style (list type)”.

[0167] Subsequently, the control unit (180) can determine at least one response style preferred by the user based on user input regarding at least one survey related to the response style received through at least one page (800). For example, the control unit (180) can receive a user's selection for the second response (902) based on the selection of one of a plurality of responses (first response (901) and second response (902)) with different response styles from a user terminal (10) on which at least one page (800) is output (second response (902)), and can determine at least one response style preferred by the user as the first response style based on the user's selection for the second response (902).

[0168] Next, the control unit (180) may provide at least one survey related to the information usage range (or at least one survey related to the information usage range) on the user terminal (10) to determine the information usage range preferred by the user. For example, as illustrated in FIGS. 9 and 10, the control unit (180) may provide at least one survey related to the information usage range through at least one page (800) based on the selection of a graphic object (e.g., next question, 903) associated with an information usage range survey start function included in at least one page (800) from the user terminal (10) after all surveys regarding response styles have been completed. In this case, the survey regarding the scope of information usage may include candidate question-response pairs configured to have a question (e.g., “How many years must pass from the date of claim for a share of pension benefits to which the statute of limitations for the share of share applies to expire?”, 910) and multiple responses (911, 912) with different scopes of information usage for said question (910). Alternatively, it may include a questionnaire configured to have a question (910) and multiple responses (or graphic objects corresponding to each of the multiple responses (911, 912)) with different scopes of information usage for said question (910). In one embodiment, among the multiple responses (911, 912), the scope of information usage that the first response (911) has (or includes) may correspond to the “second scope of information usage (detailed information type),” and the scope of information usage that the second response (912) has may correspond to the “first scope of information usage (knowledge expansion type).”

[0169] Subsequently, the control unit (180) can determine the minimum information usage range preferred by the user based on user input regarding at least one survey related to the information usage range received through at least one page (800). For example, the control unit (180) can receive a user's selection for the second response (912) based on the selection of one of a plurality of responses (first response (911) and second response (912)) with different information usage ranges from a user terminal (10) on which at least one page (800) is output (second response (912)), and can determine the at least one information usage range preferred by the user as the first information usage range according to the user's selection for the second response (912).

[0170] Furthermore, the control unit (180) can determine the user's response preference based on user input regarding a survey received through at least one page (800). That is, the control unit (180) can determine the user's response preference based on user input (or selection) regarding a survey including at least one candidate question-answer pair received through at least one page (800).

[0171] More specifically, as described above, the present invention can receive user input for each of at least one survey related to a response style and at least one survey related to an information usage scope through at least one page (800). In this regard, the control unit (180) can determine the user's response preference based on the user input for each of at least one survey related to a response style and at least one survey related to an information usage scope received through at least one page (800).

[0172] The control unit (180) can determine the user's response preference from at least one response style preferred by the user determined through at least one survey related to the response style and at least one information usage range preferred by the user determined through at least one survey related to the information usage range. That is, the control unit (180) can determine the user's response preference based on (or based on, using, etc.) the user's preferred response style determined through the results of at least one survey related to the response style and the user's preferred information usage range determined through the results of at least one survey related to the information usage range.

[0173] Furthermore, the user's response preferences (e.g., response style, scope of information usage, etc.) identified through the survey process described above may be stored in a designated storage unit (e.g., storage unit (140) or at least one memory, etc.) linked to a user account using the data generation platform. These user response preferences are continuously updated according to the user's selection (or user selection received (or collected)) made during the process of generating data according to the present invention, and the updated results may be stored and / or accumulated in a designated storage unit linked to the user account. In one embodiment, the user's response preferences may also be understood as the history information of the user (or user account).

[0174] Meanwhile, when the user's response preference is fully identified, the control unit (180) can output the identified user's response preference (or survey results regarding the response preference) on at least one page.

[0175] For example, as illustrated in FIGS. 10 and 11, the control unit (180) can output a user's response preference (1101) including at least one response style and at least one information usage range preferred by the user on at least one page (1150), based on the selection of a graphic object (e.g., “Next Question”, 913) associated with a response preference survey completion function included on at least one page (800) from the user terminal (10). That is, the control unit (180) can output a user's response preference including a judgment result (1101a) regarding the response style preferred by the user and a judgment result (1101b) regarding the information usage range preferred by the user, respectively, on at least one page (1150).

[0176] In this case, at least one page (1150) may output the user's response preference, including the preference ratio of at least one response style preferred by the user and the preference ratio of the information usage scope. For example, the judgment result (1101a) for the response style preferred by the user may include the user's preference ratio for each of the multiple response styles included in the response style (e.g., “list type 27%”, “descriptive type 43%”, “hierarchical type 30%”), and the judgment result (1101b) for the information usage scope preferred by the user may include the user's preference ratio for each of the multiple information usage scopes included in the information usage scope (e.g., “knowledge expansion type 80%”, “detailed information type 37%”, “core extraction type 33%”).

[0177] In this way, the present invention provides a user environment for identifying the response preferred by the user, thereby enabling the user to quickly and easily identify the preferred response style with just a few clicks.

[0178] Meanwhile, in the present invention, at least one processor processes at least one document and at least one user query as input to at least one artificial intelligence model (S520), and in at least one artificial intelligence model executed by at least one processor, a process of generating at least one question-response pair satisfying at least one pre-set criterion based on at least one document and at least one user query may be carried out (S530, see FIG. 5).

[0179] For example, as illustrated in FIGS. 11 and 12, the control unit (180) may provide at least one page (1250) configured to identify at least one generation flow among a data set satisfying a first criterion (a data set specialized for a domain) and a data set satisfying a second criterion (a data set generated by reflecting the user's response preference), based on the selection of a graphic object (e.g., “Next”, 1100a) associated with a function to identify the generation flow of at least one data set included in the page (1150) from the user terminal (10). Through the page (1250), the user can identify the generation flow of a question-response pair (e.g., output data (or response, 1201) generated based on the user's input data (e.g., document, query, etc., 1201)) to be included in at least one data set.

[0180] Subsequently, the control unit (180) may generate at least one data set including at least one question-response pair based on at least one document and at least one user query in response to at least one data set generation request received through the user terminal (10) on which the page (1250) is output. For example, as illustrated in FIGS. 12 and 13, when a graphic object (e.g., “Data Profile Confirmation”, 1200a) associated with a data set confirmation function included in the page (1250) is selected from the user terminal (10), the control unit (180) may output a GUI (Graphical User Interface, 1260) for receiving a data set generation request on the user terminal (10, or page (1250)). And, the control unit (180) can generate at least one data set including at least one question-response pair based on the selection of a graphic object (e.g., “Dataset creation”, 1260a) associated with a data set creation request reception function included in the GUI (1260) from the user terminal (10).

[0181] Specifically, the control unit (180, or at least one processor) can process at least one document and at least one user query as input to at least one artificial intelligence model (e.g., a large-scale language model (LLM)). And, the control unit (180) can use the artificial intelligence model to generate at least one question-response pair satisfying at least one of a first criterion and a second criterion from at least one document and at least one user query.

[0182] As described above, the at least one question-answer pair generated through the artificial intelligence model in the present invention may include at least one of at least one question-answer pair satisfying a first criterion and at least one question-answer pair satisfying a second criterion. In this case, the first criterion relates to generating a question-answer pair specialized for at least one domain, and the second criterion may relate to generating a question-answer pair that reflects the user's response preference.

[0183] That is, in the present invention, using an artificial intelligence model executed by a control unit (180), at least one of at least one question-answer pair satisfying a first criterion and at least one question-answer pair satisfying a second criterion can be generated from at least one document and at least one user query.

[0184] In this regard, let us first examine the process of generating at least one question-answer pair that satisfies the first criterion.

[0185] The control unit (180) can generate at least one question-response pair satisfying a first criterion from at least one document and at least one user query using an artificial intelligence model. As described above, the process of generating at least one question-response pair satisfying a first criterion in the present invention can be understood as a process of generating at least one question-response pair specialized in at least one domain specified from at least one of at least one document and at least one user query.

[0186] To this end, the control unit (180) can process at least one document and at least one user query as input to an artificial intelligence model. For example, the control unit (180) can configure an input prompt including a plurality of documents (604, 605, 606) received from a user terminal (10) and a plurality of user queries (604a, 604b, 604c) for at least one of the plurality of documents, and process the configured input prompt as input to an artificial intelligence model.

[0187] In this case, the artificial intelligence model can generate at least one augmented query by performing augmentation on at least one user query based on at least one document. For example, the artificial intelligence model can generate various types of augmented queries having a similar meaning to the user query by using content included in at least one and / or at least one of the received multiple documents (604, 605, 606) (e.g., content associated (or related) to the user query, paragraphs, sections, etc. associated with the user query) to perform augmentation on at least one and / or at least one of the multiple user queries (604a, 604b, 604c).

[0188] Additionally, the artificial intelligence model can generate at least one response to at least one user query and / or at least one response to at least one augmented query based on at least one document. For example, the artificial intelligence model can generate at least one response to each of the multiple user queries (604a, 604b, 604c) (multiple responses) and at least one response to at least one augmented query by using content included in at least one of the multiple documents (604, 605, 606) and / or at least one or more documents (e.g., content, paragraphs, sections, etc. associated with the user query and / or augmented query).

[0189] At least one question-answer pair satisfying the first criterion can be generated using at least one of at least one user query and at least one response to said at least one user query and at least one augmented query and at least one response to said at least one augmented query. For example, the data generation unit (160) can generate (or configure) at least one question-answer pair satisfying the first criterion by using at least one of a plurality of user queries (604a, 604b, 604c) and a response to each of the plurality of user queries (604a, 604b, 604c) obtained from an artificial intelligence model, at least one augmented query and at least one response to the at least one augmented query obtained from an artificial intelligence model.

[0190] That is, at least one question-answer pair (or question-answer pairs, multiple question-answer pairs, etc.) satisfying the first criterion may include at least one of at least one question-answer pair configured to have at least one user query and at least one response to said at least one user query, and at least one question-answer pair configured to have at least one augmented query and at least one response to said at least one augmented query.

[0191] Next, we will look at the process of generating at least one question-answer pair that satisfies the second criterion.

[0192] The control unit (180) can use an artificial intelligence model to generate at least one question-response pair satisfying a second criterion from at least one document and at least one user query. As described above, the process of generating at least one question-response pair satisfying a second criterion in the present invention can be understood as the process of generating at least one question-response pair reflecting the user's response preference from at least one document and at least one user query.

[0193] To this end, the control unit (180) can process at least one document and at least one user query as input to an artificial intelligence model. For example, the control unit (180) can configure an input prompt including a plurality of documents (604, 605, 606) received from a user terminal (10) and a plurality of user queries (604a, 604b, 604c) regarding at least one of the plurality of documents, and process the configured input prompt as input to an artificial intelligence model. At this time, the input prompt for generating a question-response pair satisfying a second criterion may further include a user's response preference (or user response preference information) and be processed as input to an artificial intelligence model. The control unit (180) can configure an input prompt including at least one of the user's response preference specified through the survey process described above, the document, and the user query, and process it as input to an artificial intelligence model. Alternatively, the control unit (180) may configure an input prompt including at least one of the user's response preference stored in a specified storage, the document, and the user query, and process it as input to the artificial intelligence model.

[0194] In this case, the artificial intelligence model can generate at least one augmented query by performing augmentation on at least one user query based on at least one document. For example, the artificial intelligence model can generate various types of augmented queries having a similar meaning to the user query by using content included in at least one and / or at least one of the received multiple documents (604, 605, 606) (e.g., content related to the user query, paragraphs, sections, etc. related to the user query) to perform augmentation on at least one and / or at least one of the multiple user queries (604a, 604b, 604c).

[0195] Additionally, the artificial intelligence model may generate at least one response reflecting (or satisfying) the second criterion for at least one user query and / or at least one response reflecting the second criterion for at least one augmented query based on at least one document. For example, the artificial intelligence model may generate at least one of a response reflecting the second criterion (multiple responses) for each of the multiple user queries (604a, 604b, 604c) and at least one response reflecting the second criterion for at least one augmented query by using content included in at least one of the multiple documents (604, 605, 606) and / or at least one document (e.g., content, paragraph, section, etc. associated with the user query and / or augmented query).

[0196] At least one question-answer pair satisfying the second criterion may be generated using at least one of at least one user query and at least one response generated by reflecting (or being made to reflect) the second criterion for said at least one user query, and at least one augmented query and at least one response generated by reflecting the second criterion for said at least one augmented query. For example, the data generation unit (160) may generate (or configure) at least one question-answer pair satisfying the second criterion by using a plurality of user queries (604a, 604b, 604c) and an AI model response generated by reflecting the second criterion for each of said plurality of user queries (604a, 604b, 604c), and at least one of at least one augmented query and an AI model response generated by reflecting the second criterion for said at least one augmented query.

[0197] That is, at least one question-answer pair (or question-answer pairs, multiple question-answer pairs, etc.) satisfying the second criterion may include at least one of at least one question-answer pair configured to have at least one user query and at least one response reflecting the second criterion for at least one user query, and at least one question-answer pair configured to have at least one augmented query and at least one response reflecting the second criterion for said at least one augmented query.

[0198] However, the at least one criterion established in the present invention may include various additional criteria in addition to the first criterion and the second criterion for generating domain-specific data and data reflecting user preferences, respectively. As an embodiment, the present invention may further include a third criterion for generating data that satisfies the first criterion and the second criterion simultaneously (or both). The present invention may generate at least one question-response pair satisfying the third criterion based on at least one document and at least one query. In this case, the process of generating a question-response pair satisfying the third criterion may also be understood as a process of generating a question-response pair that is specialized for at least one domain and reflects the user's response preference (or generating a data set containing the question-response pair).

[0199] Furthermore, in the present invention, a process of constructing at least one data set using at least one question-answer pair may be carried out (S540, see FIG. 5).

[0200] A data generation system (100) can construct at least one data set using at least one question-answer pair satisfying a first criterion and at least one question-answer pair satisfying a second criterion.

[0201] In one embodiment, the data generation system (100) can construct a first data set (410) including question-response pairs satisfying the first criterion by using question-response pairs satisfying the first criterion.

[0202] In another embodiment, the data generation system (100) can construct a second data set (420) containing question-response pairs satisfying the second criterion by using question-response pairs satisfying the second criterion.

[0203] In another embodiment, the present invention may construct at least one data set by including question-answer pairs satisfying a first criterion and question-answer pairs satisfying a second criterion in a single data set. The data generation system (100) may construct an entire data set that includes all question-answer pairs satisfying the first criterion and question-answer pairs satisfying the second criterion.

[0204] Meanwhile, the control unit (180) can store at least one data set including at least one question-response pair and / or at least one question-response pair in at least one memory, and provide (or output, display, etc.) the stored data set to the user terminal (10).

[0205] The control unit (180) may provide a first data set (410) constructed to include at least one question-answer pair satisfying a first criterion on the user terminal (10). For example, as illustrated in FIG. 14a, the control unit (180) may provide the first data set (410) containing question-answer pairs (1401, 1402, 1403) satisfying the first criterion, generated based on user input (e.g., a plurality of documents (604, 605, 606) and a plurality of user queries (604a, 604b, 604c), etc.), through at least one page (1400) output on the user terminal (10).

[0206] In this case, a query-response pair satisfying a first criterion constituting the first data set (410) may be configured such that at least one query specialized in at least one domain and at least one response to said query form a pair. For example, among the query-response pairs (1401, 1402, 1403) satisfying the first criterion, the first query-response pair (1401) may be configured such that a first query (1401a) specialized in a specific domain and a first response (1401b) to said first query (1401a) form a pair. Additionally, the second query-response pair (1402) may be configured such that a second query (1402a) specialized in a specific domain and a second response (1402b) to said second query (1402a) form a pair. Additionally, the third question-response pair (1403) may be configured such that a third question (1403a) specialized for a specific domain and a third response (1403b) to the third question (1403a) form a pair.

[0207] At this time, the control unit (180) may output to the user terminal (10) the quality evaluation result of the first data set (410) calculated according to the pre-set quality evaluation criteria together with the first data set (410). In this case, the quality evaluation result may include at least one quality evaluation result for the question-answer pairs (1401, 1402, 1403) constituting the first data set (410). For example, the page (1400) may include (or output) quality evaluation results for each of the question-answer pairs (1401, 1402, 1403) that satisfy a first criterion included in the first data set (410), along with quality evaluation results for the first question-answer pair (1401, 1402, 1403) (quality evaluation result for the first question-answer pair (1401) (e.g., “FAIL”, 1401c), quality evaluation result for the second question-answer pair (1402) (e.g., “PASS”, 1402c), and quality evaluation result for the third question-answer pair (1403) (e.g., “PASS”, 1403c). These quality evaluation results (1401c, 1402c, 1403c) are provided by at least one artificial intelligence model (or evaluator) included in the evaluation unit (170). In the model, it can be understood that it is calculated through the evaluation of each of the above question-answer pairs (1401, 1402, 1403).

[0208] Additionally, the control unit (180) may provide a second data set (420) constructed to include at least one question-answer pair satisfying a second criterion on the user terminal (10). For example, as illustrated in FIG. 14b, the control unit (180) may provide the second data set (420), which includes question-answer pairs (1411, 1412, 1413) satisfying the second criterion generated based on user input (e.g., a plurality of documents (604, 605, 606) and a plurality of user queries (604a, 604b, 604c), etc.) and the user's response preference, through at least one page (1400) output on the user terminal (10).

[0209] In this case, a question-response pair satisfying the second criterion constituting the second data set (420) may be configured such that at least one question and at least one response reflecting the user's response preference for said at least one question form a pair. For example, among the question-response pairs (1411, 1412, 1413) satisfying the second criterion, the first question-response pair (1411) may be configured such that the first question (1411a) and the first response (1411b) reflecting the user's response preference for said first question (1411a) form a pair. Additionally, the second question-response pair (1412) may be configured such that the second question (1412a) and the second response (1412b) reflecting the user's response preference for said second question (1412a) form a pair. Additionally, the third question-answer pair (1413) may be configured such that a third question (1403a) and a third response (1413b) reflecting the user's response preference for the third question (1413a) form a pair.

[0210] At this time, the control unit (180) may output to the user terminal (10) the quality evaluation result of the second data set (420) calculated according to the pre-set quality evaluation criteria together with the second data set (420). In this case, the quality evaluation result may include at least one quality evaluation result for the question-answer pairs (1411, 1412, 1413) constituting the second data set (420). For example, the page (1400) may include (or output) quality evaluation results for each of the question-answer pairs (1411, 1412, 1413) that satisfy the second criteria included in the second data set (420), along with quality evaluation results for the first question-answer pair (1411) (e.g., “PASS”, 1411c), quality evaluation results for the second question-answer pair (1412) (e.g., “PASS”, 1412c), and quality evaluation results for the third question-answer pair (1413) (e.g., “PASS”, 1413c). These quality evaluation results (1411c, 1412c, 1413c) are provided by at least one artificial intelligence model (or evaluator) included in the evaluation unit (170). In the model, it can be understood that it is calculated through the evaluation of each of the above question-answer pairs (1411, 1412, 1413).

[0211] Furthermore, the present invention can provide a user environment in which the user can perceive the generation results for question-answer pairs satisfying a first criterion included in a first data set (410) and / or question-answer pairs satisfying a second criterion included in a second data set (420) in more detail and intuitively.

[0212] For example, as illustrated in FIG. 15a, the control unit (180) may output a GUI (1510) on the user terminal (10, or page (1400)) that allows the user to recognize the generation result for the selected question-answer pair (1502) in more detail and intuitively, based on the selection of any one of the question-answer pairs (1501, 1502, 1503) that satisfy a first criterion or a second criterion included in at least one page (1500) from the user terminal (10). The GUI (1510) may be configured so that the user can intuitively recognize at least one of the generation result and the quality evaluation result (e.g., “AI quality evaluation”) for the selected question-answer pair (1502).

[0213] Here, the control unit (180) may provide a user environment that allows the user to identify errors in the response to a query in a document by comparing it with the content of the document, based on the selection of at least one graphic object (e.g., “View Report”, 1510a) included in the GUI (1510) from the user terminal. For example, as illustrated in FIG. 15b, the control unit (180) may provide a user environment that allows the user to identify errors in the output data (or response, 1502b) generated through the input data (1502a) and the content included in at least one specific document (e.g., “National Pension Service Sample Data.pdf”) included in the input data (1502a) used as input in the process of generating the query-response pair (1502) having a specific quality (e.g., second quality) among a plurality of qualities (first quality (PASS), second quality (FAIL)). That is, the control unit (180) can provide a user environment that can identify errors in the response to a query of the specific document by comparing it with the content included in the specific document. This can be provided through at least one page (1500) output to the user terminal (10) in the form of a report.

[0214] Meanwhile, in the present invention, at least one target model for learning is specified, and the specified target model for learning can be trained using at least one data set stored in at least one memory.

[0215] For example, as illustrated in FIGS. 16 and 17, the control unit (180) may output a GUI (1710) configured to receive a fine-tuning request on the user terminal (10) or the page (1700) based on the selection of a graphic object (1700a) associated with a GUI output function for a fine-tuning request included in the page (1700) from the user terminal (10) on which at least one page (1700) has been output. Through the GUI (1710), the user may input at least one of the following: model type (e.g., “EXAONE-3.5”), model size (e.g., “7.8B”), model evaluation method (e.g., “Evaluation Set FAQ 30 items, evaluation method is cross-evaluation by 4 field personnel, evaluation criterion is accuracy”), requesting organization (e.g., “AX Department”), intended use and expected effect (e.g., “Employee training and improvement of productivity in public service operations”). Subsequently, the control unit (180) may output the GUI (1710) included Based on the selection of a graphic object (1710a) associated with a fine-tuning request receiving function, a fine-tuning request including at least one user input for fine-tuning can be received.

[0216] In this case, at least one learning target model may be specified based on a user selection included in a fine-tuning request. For example, the control unit (180) may specify one of the multiple learning target models (or multiple learning target artificial intelligence models) selected by the user terminal (10) (e.g., “EXAONE-3.5”) as the learning target (i.e., fine-tuning target) model.

[0217] Furthermore, the control unit (180) can automatically perform at least one generated data set in conjunction with the fine-tuning of the target model. The control unit (180) can automatically perform the first data set (410) and the second data set (420) in conjunction with the fine-tuning of the target model, respectively.

[0218] More specifically, the control unit (180) can perform fine-tuning on at least one learning target model using at least one of question-answer pairs (1401, 1402, 1403) satisfying a first criterion included in the first data set (410) and question-answer pairs (1411, 1412, 1413) satisfying a second criterion included in the second data set (420). Alternatively, the control unit (180) may perform fine-tuning on a model to be trained so that the model to be trained is specialized for each of the first data set (410) and the second data set (420) by using a first data set (410) containing question-answer pairs (1401, 1402, 1403) satisfying a first criterion and a second data set (420) containing question-answer pairs (1411, 1412, 1413) satisfying a second criterion, respectively.

[0219] In one embodiment, the process of fine-tuning a model to be trained through a first dataset (410) may be a process of fine-tuning a model to be trained with a first dataset (410) containing question-answer pairs (1401, 1402, 1403) satisfying a first criterion so that the model to be trained is specialized in at least one domain. In this case, at least one parameter of the model to be trained (or at least one parameter included in the model to be trained) may be adjusted (or tuned) so as to be specialized in at least one domain.

[0220] In another embodiment, the process of fine-tuning the model to be trained through the first data set (410) may be a process of fine-tuning the model to be trained with a second data set (420) containing question-response pairs (1411, 1412, 1413) satisfying the second criterion so that the model to be trained generates a response reflecting the second criterion for at least one query. In this case, at least one parameter of the model to be trained may be adjusted to generate a response that reflects the user's response preference.

[0221] In another embodiment, the fine-tuning process of a model to be trained using both the first data set (410) and the second data set (420) may be a process of fine-tuning the model to be trained with each of the first data set (410) and the second data set (420) (or separately) such that the model to be trained is specialized for at least one domain and generates a response reflecting a second criterion for a query related to said domain. In this case, at least one parameter of the model to be trained may be adjusted to be specialized for at least one domain and to generate a response reflecting the user's response preference.

[0222] Not limited thereto, the present invention can provide a user environment that allows a user to build a customized model by selectively utilizing desired data among the generated data.

[0223] In one embodiment, the control unit (180) may receive a user input requesting fine-tuning of at least one target model using at least one of a first data set (410) and a second data set (420) through a user terminal (10, or at least one page). Then, the control unit (180) may specify at least one data set among the first data set (410) and the second data set (420) to be used for fine-tuning based on the user input received through the user terminal (10). In this case, if the specified data set includes the first data set (410), fine-tuning of the target model may be performed using the first data set (410). Alternatively, if the specified data set includes the second data set (420), fine-tuning of the target model may be performed using the first data set (410). Alternatively, if the specified data set includes a first data set (410) and a second data set (420), fine-tuning can be performed on the model to be trained using the first data set (410) and the second data set (420), respectively.

[0224] Subsequently, when training of the model to be trained is completed, the control unit (180) may obtain a trained model (or a fine-tuned model) through at least one data set (a first data set (410), a second data set (420)). The fine-tuned model may be specialized for at least one domain or configured to generate a response that reflects the user's response preference. Alternatively, the fine-tuned model may be configured to satisfy a first criterion while simultaneously generating a response that satisfies a second criterion for a query related to the at least one domain. This may mean that the trained model is specialized for at least one domain while simultaneously generating a response that reflects the user's response preference for a query related to the at least one domain.

[0225] Finally, the control unit (180) stores the fine-tuned model and at least one data set used for training the fine-tuned model in a specified storage (e.g., storage unit (140) or at least one memory) in association, and can provide the fine-tuned model through at least one page associated with the user terminal (10) and / or the data generation platform and / or at least one user account registered with the data generation platform. For example, the fine-tuned model may be associated with a user account and stored in the specified storage together with at least one data set used for fine-tuning.

[0226] Meanwhile, the control unit (180) can output the response quality of a model fine-tuned with at least one data set to the user terminal (10). Here, the fine-tuned model may include a model fine-tuned with at least one of the first data set (410) and the second data set (420) generated through the process described above.

[0227] For example, as illustrated in FIG. 18, the control unit (180) may output at least one page (1900) on the user terminal (10) to enable checking the response quality (or response performance) of the fine-tuned model in response to a user's request to check the response quality (or response performance) of the fine-tuned model. The page (1900) may be configured to evaluate and output the response quality of the fine-tuned model by searching for at least one document related to (or associated with) the user query when the user inputs a user query related to at least one domain through a query receiving area (1910) included in the page (1900). The control unit (180) can receive a user query (e.g., “Input question”, 1911a) associated with a query receiving function included in the page (1900, or query receiving area (1910)) from the user terminal (10) based on the selection of a graphic object (e.g., “Input question”, 1911a) associated with at least one domain (e.g., “I am a university lecturer working less than 60 hours a month, and I signed an employment contract for the first semester (April 25 - June 25), but I was told that I gave lectures in March, May, and June, but not in April. The reason given is that there are several lecturers, so there are months when I am not teaching! In this case, I would like to inquire whether this falls under the reason for 3 consecutive months of work and is subject to acquisition reporting, and how the standard monthly income is calculated.”, 1911).

[0228] Subsequently, the control unit (180) can use the data processing unit (150) to search for at least one document related to (or associated with) the user query (1911). The data processing unit (150) can use a document search module to search for documents related to the user query (1911) from a previously specified storage (e.g., storage unit (140) or at least one memory) and / or various sources. The control unit (180) processes the searched document and the user query regarding the searched document as inputs to the model prior to fine-tuning (or the model prior to fine-tuning, the first model, etc.) and the model after fine-tuning (or the fine-tuned model, the second model, etc.), respectively, and can obtain a response to the user query from the model prior to fine-tuning and the fine-tuned model, respectively.

[0229] At this time, as an example embodiment, when a document (1912) related to a user query (1911) is searched, as illustrated in FIG. 19, the control unit (180) may identify related content (or related content) related to the user query among the contents included in the searched document (1912), and highlight (or display, output, provide, etc.) the identified related content on a part of the page (1900).

[0230] Furthermore, the control unit (180) can use the evaluation unit (170) to evaluate the quality of the first response of the model prior to fine-tuning and the second response of the model prior to fine-tuning for a user query (1911), respectively. At least one evaluator model included in the evaluation unit (170) can evaluate the quality of the first response of the model prior to fine-tuning and the second response of the model prior to fine-tuning for a user query (1911), respectively, based on the retrieved document (1912).

[0231] Finally, the control unit (180) can output to the user terminal at least one of the response quality of the model before fine-tuning for the query and the response quality of the model after fine-tuning for the query. For example, as illustrated in FIG. 20, the control unit (180) can output the first response quality (1921) of the model before fine-tuning and the second response quality (1922) of the model after fine-tuning for the user query (1911) evaluated through the evaluation unit (170) on the page (1900) output to the user terminal (10), respectively. That is, the user can be provided with an environment in which they can intuitively compare the response quality of the model before fine-tuning and the response quality of the model after fine-tuning.

[0232] Meanwhile, as seen above, at least one data set generated in response to a user's request can be stored in at least one memory of the data generation system (100) in association with a user account.

[0233] A data generation system (100) can perform clustering on question-answer pairs included in at least one dataset based on various clustering criteria. Here, clustering may mean grouping data with similar characteristics into one cluster (or cluster or group) and separating data with different characteristics into different clusters.

[0234] In the present invention, the criteria for clustering can be set in various ways. For example, the clustering criteria may include at least one of i) semantic similarity criteria (e.g., clustering based on intent and cosine similarity in the sentence embedding space), ii) domain keyword and conceptual term-based criteria (e.g., classification based on the frequency of keyword occurrence or concepts or detailed categories used in a specific professional domain), iii) response type-based criteria (e.g., morphological characteristics of the response (response style or scope of information usage, etc.)), iv) domain difficulty and style-based criteria (e.g., clustering based on the difficulty of the query (basic, intermediate, advanced) or the technical level of the response (e.g., overview level vs. research level)), and v) query type-based criteria (e.g., clustering based on the purpose of the query (definitive type (“what is~?”), comparative type (“the difference between~ and~?”), procedural type (“how to do~?”), causal type (“why does~ happen?”), etc.)). However, this is merely one example, and it is obvious that the examples included in the clustering criteria of the present invention may include various additional examples besides those mentioned above.

[0235] In one embodiment, let us assume that at least one specific data set includes question-and-answer pairs specialized for a specific domain (e.g., manufacturing process). In this case, the data generation system (100) may cluster (or classify) at least one question-and-answer pair containing a first keyword (e.g., cause of defect (bubble, crack, shrinkage, contamination, delamination, etc.)) among the question-and-answer pairs included in the specific data set (i.e., keyword analysis result) into a first group (or cluster). Additionally, the data generation system (100) may cluster at least one question-and-answer pair containing a second keyword (e.g., process optimization (temperature, pressure, speed, concentration, pH adjustment, etc.)) among the question-and-answer pairs into a second group. Additionally, the data generation system (100) can cluster at least one question-answer pair containing a third keyword (e.g., safety and maintenance (explosion, leak, chemical management, periodic inspection, etc.) among the question-answer pairs into a third group.

[0236] Subsequently, the data generation system (100) may store at least one group and / or at least one group that has been clustered within a specific data set according to clustering criteria in at least one memory associated with a user account. For example, the data generation system (100) may store a plurality of groups (first group, second group, third group) obtained as a result of performing clustering on a specific data set in at least one memory associated with a user account.

[0237] In this case, the present invention may provide a user environment in which a user can selectively utilize a desired group among a plurality of groups to build a customized model. For example, the control unit (180) may receive user input requesting fine-tuning of at least one target model using at least one of a plurality of groups through a user terminal (10, or at least one page). Then, based on the user input received through the user terminal (10), the control unit (180) may specify at least one group among the plurality of groups to be used for fine-tuning, and perform fine-tuning on the target model using the specified group.

[0238] In this way, the present invention provides a user environment in which a user can intuitively and selectively utilize desired data within a data set by performing clustering on question-response pairs included in at least one data set. That is, the user can be provided with an environment in which not only a customized model specialized for a specific domain but also a model specialized in detail within a specific domain can be constructed.

[0239] Meanwhile, according to another embodiment of the data generation method and system according to the present invention, the data generation system (100) according to the present invention may provide a user with a function to build a multi-domain model by selecting and / or combining clusters corresponding to at least one and / or at least one (multiple) domains.

[0240] In this case, the data generation unit (160) can integrate question-response pairs extracted from each of a plurality of different (or distinct) domains into a single model training pipeline and, by performing fine-tuning on the model to be trained based on the combined cluster (or dataset), can build a user-customized model that solves interdependent problems between a plurality of different domains.

[0241] In one embodiment, a user can use a data generation system (100) according to the present invention to combine a first cluster (e.g., manufacturing process domain cluster) containing question-answer pairs corresponding to (or about, related to, etc.) a first domain (e.g., manufacturing process) and a second cluster (e.g., medical domain cluster) containing question-answer pairs corresponding to a second domain (e.g., medical), and build a multi-domain model that addresses safety and / or medical risk assessment issues that may occur in a manufacturing process environment using (or based on) the combined cluster.

[0242] In another embodiment, the user can use the data generation system (100) according to the present invention to combine a third cluster (e.g., a financial domain cluster) containing question-answer pairs corresponding to a third domain (e.g., finance) and a fourth cluster (e.g., an IT domain cluster) containing question-answer pairs corresponding to a fourth domain (e.g., an IT domain), and use the combined cluster to build a multi-domain model that solves complex domain problems such as security threat detection and / or risk analysis in a financial IT system.

[0243] In this way, the present invention automatically generates a data set that reflects a combined structure between multiple domains as well as a single domain, and thereby enables the securing of a high-quality, user-customized model based on multiple domains.

[0244] In addition, according to another embodiment of the data generation method and system according to the present invention, the data generation system (100) according to the present invention can provide real-time clustering and model update functions to the user.

[0245] In one embodiment, the data generation system (100) performs clustering in real time whenever a new user question-response pair is added (or input), and can automatically update (or fine-tune) at least one learning target model (e.g., a model stored in association with a user account) based on the clusters according to the clustering results.

[0246] In this way, the present invention can provide a user-customized model that responds in real time to changes in user input and is continuously and progressively improved.

[0247] Meanwhile, according to another embodiment of the data generation method and system according to the present invention, the data generation system (100) according to the present invention may further include a function to visually represent and provide to a user the relationship between at least one and / or at least one (multiple) clusters generated.

[0248] The data generation system (100) can analyze similarity, correlation, influence, and / or distance (metric) information between clusters and provide it to the user in various forms such as graph-based visualization, network diagrams, and node-edge structures.

[0249] In one embodiment, the data generation system (100) can automatically generate a network diagram composed of clusters as nodes and edges with similarity values ​​between clusters applied as weights. Through this, the user can intuitively check, through a visualized screen, how closely a specific cluster is connected to other clusters or whether it is independent.

[0250] In another embodiment, the data generation system (100) may provide a user environment that can visually identify changes in dynamic cluster relationships associated with a real-time update function by providing a cluster structure that changes over time in the form of an animation.

[0251] As described above, according to the data generation system and control method of the present invention, domain-specific question-response pairs and question-response pairs reflecting the user's response preferences can be automatically generated using an artificial intelligence model, and a target model for training can be fine-tuned based on the generated question-response pairs. Through this, the present invention can secure a model that is specialized for the domain and simultaneously generates user-customized responses according to the user's response preferences.

[0252] Furthermore, according to the data generation system and control method of the present invention, at least one of high-quality domain-specific data and data reflecting user preferences can be automatically generated using an artificial intelligence model, and a target model for training can be precisely tuned based on this. Through this, users can generate various data with minimal manpower, cost, and time, and receive a model built and tuned to suit their needs. In other words, users can receive a model that generates customized responses based on user preferences while simultaneously improving response accuracy and performance in the domain.

[0253] Furthermore, according to the data generation system and control method of the present invention, a user environment can be provided that automates everything from the generation of domain-specific data and data reflecting user preferences to model training. Through this, users can easily and simply generate desired data quickly and safely, thereby increasing data productivity and efficiency, and automatically build a customized, optimized model.

[0254] Furthermore, according to the data generation system and control method of the present invention, data generation results generated by reflecting user input and preferences, and training and performance evaluation results of a model fine-tuned with the generated data can be visualized and output. That is, by enhancing data understanding based on visualization in the present invention, the user is provided with an environment in which they can intuitively recognize the generated data and the training and performance evaluation results of the fine-tuned model.

[0255] Meanwhile, the present invention described above can be implemented based on a quantum computer. The present invention implemented based on a quantum computer may include a qubit-based quantum processor and quantum memory, and may include software and hardware interfaces optimized for quantum computation.

[0256] Quantum processors in quantum computers utilize qubits to efficiently process complex operations through parallel computation, quantum entanglement, and quantum superposition, which cannot be performed by the binary bits of classical computers. Quantum processors process data using quantum gates and can provide exponential speed improvements for specific problems.

[0257] Meanwhile, the present invention described above can be implemented as a program that is executed by one or more processes on a computer and can be stored on a computer-readable medium (or recording medium).

[0258] Furthermore, the present invention described above can be implemented as computer-readable code or instructions on a medium on which a program is recorded. That is, the present invention can be provided in the form of a program.

[0259] Meanwhile, computer-readable media include all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0260] Furthermore, the computer-readable medium may be a server or cloud storage that includes a storage and is accessible to an electronic device via communication. In this case, the computer may download the program according to the present invention from the server or cloud storage via wired or wireless communication.

[0261] A computer program may reach the system (100) through various suitable transmission mechanisms. The transmission mechanism may be, for example, a computer-readable storage medium, a computer program product, a memory device, a recording medium such as a CD-ROM or DVD, or a product that tangibly embodies the computer program. The transmission mechanism may be a signal configured to reliably transmit the computer program through air or an electrical connection. The system (100) may propagate or transmit the computer program as a computer data signal.

[0262] Furthermore, references to 'computer-readable storage media,' 'computer program products,' 'computer programs embodied in a tangible form,' etc., or to 'controller,' 'computer,' 'processor,' etc., should be understood to include not only computers with various architectures such as single / multi-processor architectures and sequential (Von Neumann) / parallel architectures, but also specialized circuits such as Field-Programmable Gate Arrays (FPGAs), Application Specific Circuits (ASICs), signal processing units, and other devices. References to computer programs, instructions, code, etc., should be understood to include software for programmable processors or firmware, such as programmable content for hardware devices, whether it is instructions for a processor or configuration settings for a fixed-function device, gate array, or programmable logic device.

[0263] Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, namely a CPU (Central Processing Unit), and no special limitations are placed on its type.

[0264] Meanwhile, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention.

Claims

Claim 1 A method for controlling a data generation system, comprising: receiving at least one user query for at least one document from a user terminal, in a method performed by a computer; a step in which at least one processor processes the at least one document and the at least one user query as input to at least one artificial intelligence model; a step in which at least one question-response pair satisfying at least one pre-set criterion based on the at least one document and the at least one user query is generated in the at least one artificial intelligence model executed by the at least one processor; and a step of constructing at least one data set using the at least one question-response pair, wherein the at least one pre-set criterion includes at least one of a first criterion related to the generation of a domain-specific question-response pair and a second criterion related to the generation of a question-response pair reflecting a user's response preference, and wherein the step of generating the at least one question-response pair is characterized by generating at least one question-response pair satisfying at least one of the first criterion and the second criterion based on the at least one document and the at least one user query. Claim 2 A control method for a data generation system according to claim 1, wherein in the step of constructing the at least one data set, the at least one data set is constructed using the at least one question-answer pair satisfying at least one of the first criterion and the second criterion. Claim 3 A method for controlling a data generation system according to claim 2, wherein the at least one question-answer pair comprises at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion, and in the step of generating the at least one question-answer pair, at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion is generated from the at least one document and the at least one user query using the at least one artificial intelligence model performed by the at least one processor. Claim 4 A method for controlling a data generation system according to claim 3, wherein the step of generating at least one question-answer pair comprises: a step of generating at least one question-answer pair satisfying the first criterion from the at least one document and the at least one user query using the at least one artificial intelligence model performed by the at least one processor; and a step of generating at least one question-answer pair satisfying the second criterion from the at least one document and the at least one user query using the at least one artificial intelligence model performed by the at least one processor. Claim 5 A method for controlling a data generation system, wherein, in claim 4, the step of generating at least one question-response pair satisfying the first criterion comprises: the step of the at least one processor processing the at least one document and the at least one user query as inputs to at least one artificial intelligence model; the step of generating at least one augmented query by performing augmentation on the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; the step of generating at least one response to the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; and the step of generating at least one response to the at least one augmented query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor. Claim 6 A control method for a data generation system according to claim 5, wherein in the step of generating at least one question-response pair satisfying the first criterion, at least one of the at least one user query and at least one response to the at least one user query and at least one augmented query and at least one response to the at least one augmented query is used to generate at least one question-response pair satisfying the first criterion. Claim 7 A method for controlling a data generation system according to claim 6, wherein at least one question-answer pair satisfying the first criterion comprises at least one of at least one question-answer pair configured to have at least one user query and at least one response to the at least one user query, and at least one question-answer pair configured to have at least one augmented query and at least one response to the at least one augmented query. Claim 8 A control method for a data generation system according to claim 4, wherein the step of generating at least one question-response pair satisfying the first criterion is the step of generating at least one question-response pair specialized in at least one domain specified from at least one of the at least one document and at least one user query. Claim 9 A method for controlling a data generation system according to claim 8, wherein at least one question-response pair satisfying the first criterion is configured such that at least one question specialized in the at least one domain and at least one response to the question form a pair. Claim 10 A method for controlling a data generation system according to claim 4, wherein the step of generating at least one question-response pair satisfying the second criterion comprises: the step of the at least one processor processing the at least one document and the at least one user query as inputs to at least one artificial intelligence model; the step of generating at least one augmented query by performing augmentation on the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; the step of generating at least one response reflecting the second criterion for the at least one user query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor; and the step of generating at least one response reflecting the second criterion for the at least one augmented query based on the at least one document in the at least one artificial intelligence model executed by the at least one processor. Claim 11 A control method for a data generation system according to claim 10, wherein in the step of generating at least one question-response pair satisfying the second criterion, at least one of the at least one user query and at least one response generated by reflecting the second criterion for the at least one user query and at least one response generated by reflecting the second criterion for the at least one user query and at least one augmented query and at least one response generated by reflecting the second criterion for the at least one augmented query is used to generate at least one question-response pair satisfying the second criterion. Claim 12 A method for controlling a data generation system according to claim 11, wherein at least one question-answer pair satisfying the second criterion comprises at least one of the following: at least one question-answer pair configured to have at least one user query and at least one response reflecting the second criterion for the at least one user query; and at least one question-answer pair configured to have at least one augmented query and at least one response reflecting the second criterion for the at least one augmented query. Claim 13 A method for controlling a data generation system, wherein, in claim 4, the step of generating at least one question-response pair satisfying the second criterion is the step of generating at least one question-response pair reflecting the user's response preference from the at least one document and the at least one user query. Claim 14 A method for controlling a data generation system according to claim 13, wherein at least one question-response pair satisfying the second criterion is configured such that at least one question and at least one response reflecting the user's response preference for the at least one question form a pair. Claim 15 A method for controlling a data generation system according to claim 2, further comprising the step of storing the at least one data set in at least one memory, wherein the at least one question-answer pair comprises at least one of the at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion. Claim 16 A method for controlling a data generation system according to claim 15, further comprising: a step of specifying at least one learning target model; and a step of training the at least one learning target model using the at least one data set stored in the at least one memory, wherein the at least one data set is constructed to include at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion. Claim 17 A control method for a data generation system according to claim 15, wherein in the step of training at least one learning target model, fine-tuning is performed on the at least one learning target model using at least one of at least one question-answer pair satisfying the first criterion and at least one question-answer pair satisfying the second criterion. Claim 18 A data generation system comprising a memory configured to store executable instructions and one or more processors configured to perform operations by executing one or more instructions, wherein the system receives at least one user query regarding at least one document from a user terminal, processes the at least one document and the at least one user query as input to at least one artificial intelligence model, generates at least one question-response pair satisfying at least one pre-set criterion from the at least one document and the at least one user query using the at least one artificial intelligence model, constructs at least one data set using the at least one question-response pair, wherein the at least one pre-set criterion includes at least one of a first criterion related to the generation of a domain-specific question-response pair and a second criterion related to the generation of a question-response pair reflecting a user's response preference, and in generating the at least one question-response pair, generates at least one question-response pair satisfying at least one of the first criterion and the second criterion based on the at least one document and the at least one user query. Claim 19 A program that is executed by one or more processes in an electronic device and stored on a computer-readable recording medium, wherein the program comprises: receiving at least one user query regarding at least one document from a user terminal; processing the at least one document and the at least one user query as input to at least one artificial intelligence model; and using the at least one artificial intelligence model to generate at least one question-response pair satisfying at least one pre-set criterion from the at least one document and the at least one user query. A program stored on a computer-readable recording medium, comprising instructions for performing the step of constructing at least one data set using at least one question-answer pair, wherein at least one criterion set in advance includes at least one of a first criterion related to the generation of a domain-specific question-answer pair and at least one of a second criterion related to the generation of a question-answer pair reflecting a user's response preference, and wherein, in generating the at least one question-answer pair, at least one question-answer pair satisfying at least one of the first criterion and the second criterion is generated based on at least one document and at least one user query.

Citation Information

Patent Citations

  • Apparatus for searching data using index and method for using the apparatus

    KR1020150096848A

  • Method, system and computer program for artificial intelligence answer

    KR1020200036086A

  • Architecture for generating QA pairs from contexts

    KR1020220063476A

  • Generation of synthetic question-answer pairs using a document classifier and classification explainer

    US20240256782A1

  • Method and apparatus self-training of machine reading comprehension to improve domain adaptation

    KR1020220142027A