Method for revising document framework based on instruction system therefor

Abstract

An instruction-based method for modifying a document framework and a system for the same are provided. An instruction-based method for modifying a document framework according to one embodiment of the present disclosure may include, in a method performed by a computing system, receiving a framework modification request from a user terminal requesting modification of the framework of a first document—the framework modification request includes an instruction identifying a reference target resource—analyzing the framework of the reference target resource using an AI model and generating a template prompt based on the framework of the reference target resource; searching for a chunk among chunks related to the first document stored in a knowledge base that has a similarity to a component included in the template prompt greater than or equal to a threshold value and obtaining an input chunk; and inputting the input chunk, the template prompt, and the framework modification request into a pre-trained Large Language Model and generating a draft content in which the framework of the first document is modified.

Description

Technology Field

[0001] The present disclosure relates to a method and system for modifying an instruction-based document framework. More specifically, the present disclosure relates to a method and system for modifying the framework of a document, such as its format, style, and logical structure, according to a user's request. Background Technology

[0002] Recently, various AI assistant services have been utilizing user data to provide functions such as document creation, summarization, and question-and-answer. In particular, artificial intelligence systems based on Large Language Models (LLM) operate by transmitting user input data to a server for analysis and generating results. While these systems often employ Retrieval-Augmented Generation (RAG) techniques to enhance response accuracy based on external resources without the AI ​​undergoing self-learning, most adopt a method of processing and storing data on a central server.

[0003] These existing technologies raise security concerns because user documents are transmitted to external servers and can be stored temporarily or permanently. In particular, for content containing legal documents, internal corporate data, and sensitive personal information, users experience persistent anxiety regarding the possibility of data leakage. Consequently, there is a growing demand for technologies that provide document creation capabilities while processing user data locally to safely protect it without external leakage. Prior art literature

[0004] Published Patent Application No. 10-2025-0081316 (June 5, 2025) The problem to be solved

[0005] The technical problem to be solved through some embodiments of the present disclosure is to provide a method and system for modifying the framework of a document, such as its format, style, and logical structure, according to a user's request.

[0006] The technical problems of the present disclosure are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by a person skilled in the art of the present disclosure from the description below. means of solving the problem

[0007] An instruction-based document framework modification method according to some embodiments of the present disclosure for solving the aforementioned technical problem may include, in a method performed by a computing system, receiving a framework modification request from a user terminal requesting modification of the framework of a first document—the framework modification request includes an instruction identifying a reference target resource—analyzing the framework of the reference target resource using an AI model and generating a template prompt based on the framework of the reference target resource; searching for a chunk among chunks related to the first document stored in a knowledge base that has a similarity to a component included in the template prompt greater than or equal to a threshold value and obtaining an input chunk; and inputting the input chunk, the template prompt, and the framework modification request into a pre-trained Large Language Model and generating a draft content in which the framework of the first document is modified. Brief explanation of the drawing

[0008] FIG. 1 is a system configuration diagram for explaining the configuration and operation of a document processing system according to some embodiments of the present disclosure. FIG. 2 is a diagram illustrating the fields of data stored in a database included in a knowledge base according to some embodiments of the present disclosure. FIG. 3 is a diagram showing the UI configuration of a document creation system displayed on a user terminal according to some embodiments of the present disclosure. FIG. 4 is a diagram showing an example of a UI configuration of a document creation system displayed on a user terminal according to some embodiments of the present disclosure. FIG. 5 is a flowchart illustrating the operation of an instruction-based document framework modification method according to some embodiments of the present disclosure. FIG. 6 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 7 is a drawing for illustrating an example of a formal prompt according to some embodiments of the present disclosure. FIG. 8 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 9 is a drawing for illustrating an example of a logic structure prompt according to some embodiments of the present disclosure. FIG. 10 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 11 is a drawing for illustrating an example of a style prompt according to some embodiments of the present disclosure. FIG. 12 is a drawing for illustrating an example of a template prompt according to some embodiments of the present disclosure. FIG. 13 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 14 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 15 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 16 is a drawing for illustrating a method of generating draft content according to some embodiments of the present disclosure. FIG. 17 illustrates an exemplary computing device capable of implementing systems according to some embodiments of the present disclosure. Specific details for implementing the invention

[0009] Hereinafter, various embodiments of the present disclosure will be described in detail with reference to the attached drawings. The advantages and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the attached drawings. However, the technical concept of the present disclosure is not limited to the following embodiments but can be implemented in various different forms. The following embodiments are provided merely to complete the technical concept of the present disclosure and to fully inform those skilled in the art of the scope of the present disclosure, and the technical concept of the present disclosure is defined only by the scope of the claims.

[0010] In describing the various embodiments of the present disclosure, if it is determined that a detailed description of related known configurations or functions could obscure the essence of the present disclosure, such detailed description is omitted.

[0011] Unless otherwise defined, terms used in the following embodiments (including technical and scientific terms) may be used in a meaning commonly understood by those skilled in the art to which this disclosure pertains, but this may vary depending on the intent of those skilled in the art, case law, the emergence of new technology, etc. The terms used in this disclosure are for describing the embodiments and are not intended to limit the scope of this disclosure.

[0012] In the following embodiments, singular expressions include plural concepts unless the context clearly specifies them as singular. Additionally, plural expressions include singular concepts unless the context clearly specifies them as plural.

[0013] In addition, terms such as first, second, A, B, (a), (b), etc. used in the following embodiments are used merely to distinguish one component from another, and the essence, order, or sequence of the said component is not limited by such terms.

[0014] Hereinafter, various embodiments of the present disclosure will be described in detail with reference to the attached drawings.

[0015] Hereinafter, with reference to FIG. 1, the configuration and operation of a document processing system according to some embodiments of the present disclosure will be described. FIG. 1 is a system configuration diagram for explaining the configuration and operation of a document processing system according to some embodiments of the present disclosure.

[0016] Referring to FIG. 1, a document processing system may be configured to include a document creation system (1), a user terminal (30), and an LLM server (40). The document creation system (1) may be configured to include a service server (10) and a knowledge base (20) including a first database (21), a second database (22), and a third database (23). However, the scope of the present disclosure is not limited thereto. In some cases, the document processing system may be configured to include additional modules / devices / systems not shown in FIG. 1. Alternatively, the document processing system may be configured in a form in which at least some of the components (1 to 40) shown in FIG. 1 are excluded.

[0017] The user terminal (30) may be a terminal used by a user of the document processing system. The user may input requests for the creation or modification of specific documents through the user terminal (30). The user terminal (30) may transmit various user requests to the service server (10).

[0018] The service server (11) can automatically create a document, modify a created document, or modify the framework of a created document according to a user's request. The service server (11) can search for data that needs to be referenced to fulfill the user's request in the knowledge base (20) using a Retrieval-Augmented Generation (RAG) technique.

[0019] The knowledge base (20) may be configured to include a first database (21), a second database (22), and a third database (23).

[0020] The first database (21) can store chunks corresponding to each semantic unit by parsing the resources (e.g., document files, web links, folders, etc.) uploaded by the user according to predefined criteria. That is, the first database (21) may be a database that stores data in which preprocessing is performed primarily on the uploaded resources by semantic unit in response to the user's resource upload request.

[0021] The second database (22) can store data mapping directory information stored in the knowledge base (20) for each of the above chunks. That is, the second database (22) can store information regarding which directory in the storage stack allocated to the document creation system (1) the data for which preprocessing has been performed on the uploaded resource is stored. When a user creates a document creation request, etc. by specifying a specific document, the service server (10) can quickly search for the specific document by using the directory information stored in the second database (22).

[0022] The service server (10) can randomly encrypt the chunk and store information regarding the encrypted chunk in the second database (22). By performing encryption in this way, even if the service server (10) inputs a document creation request, etc., to the LLM server (40) in accordance with the user's request, the original data regarding the uploaded document is not input into the LLM. Therefore, there is an advantage in that security for the resources uploaded by the user can be maintained.

[0023] The third database (23) may be a database that stores semantic embedding vectors that vectorize each of the chunks for each of the chunks. That is, the third database (23) may store embedding vectors for semantic units parsed from a specific document. The service server (10) may use the semantic embedding vectors for each of the chunks stored in the third database (23) to search for chunks corresponding to semantic embedding vectors whose similarity to the embedding vector corresponding to the user's request is greater than or equal to a threshold value.

[0024] However, the scope of the present disclosure is not limited to the above, and the knowledge base (20) may be configured to include other types of databases in addition to the databases (21, 22, 23).

[0025] The service server (10) has the advantage of being able to more quickly obtain data that needs to be referenced from among the data stored in the knowledge base (20) in response to the user's request by searching the databases (21, 22, 23) as needed according to the user's request.

[0026] Hereinafter, with reference to FIG. 2, fields of data stored in a database included in a knowledge base according to some embodiments of the present disclosure will be described. FIG. 2 is a diagram for describing fields of data stored in a database included in a knowledge base according to some embodiments of the present disclosure.

[0027] As previously explained, the knowledge base (20) may include a first database (21), a second database (22), and a third database (23).

[0028] As previously explained, the knowledge base (20) may include a first database (21), a second database (22), and a third database (23).

[0029] Referring to FIG. 2, the field (21a) of the data stored in the first database (21) may include resource identification information, chunk identification information, text information, and location information. However, the field (21a) may include other information in addition to the information listed above. Resource identification information is an identifier of the entire original document, chunk identification information is unique identification information for a chunk unit, text information is original text data of the corresponding chunk, and location information may be location information (e.g., page, line number, etc.) occupied by the chunk within the document. The service server (10) stores the data parsed in chunk units in the first database (21) while preserving location information according to the document structure, thereby maintaining the logical order of input chunks, etc., when inputting into a large-scale language model.

[0030] The field (22b) of the data stored in the second database (22) may include metadata including keywords, directory structure information, and semantic embeddings for each chunk. The chunk identification information may be unique identification information at the chunk level linked to the first database (21), the keyword may be a subject word extracted from the corresponding chunk, the directory information may be information regarding the folder path to which the corresponding chunk belongs, and the semantic embedding vector may be embedding information of the chunk linked to the third database (23). When the service server (10) acquires a search target and performs a search, it can acquire the search target more quickly by filtering chunks according to specific keywords or folder paths, or by performing a similarity search based on semantic embeddings.

[0031] The second database (22) can be linked with the first database (21) through chunk identification information and linked with the third database (23) through semantic embedding vector information. Accordingly, the service server (10) can filter keywords that match the search target or chunks that have semantic embeddings with a similarity to the search target that is greater than or equal to a threshold value by searching the second database (22) for the search target. Accordingly, by including directory information of each chunk in the field (22b) of the data included in the second database (22), the document creation system (1) can obtain document preprocessing data more quickly.

[0032] The field (23a) of the data stored in the third database (23) may include information regarding chunk identification information and semantic embedding vectors of the corresponding chunk. The service server (10) may perform context-based search (RAG) by embedding the user's document creation request and performing a similarity-based comparison with the third database (23).

[0033] According to the present embodiment, by separating the knowledge base (20) into semantic unit text, metadata, and semantic vector and storing them in each database (21, 22, 23), search accuracy and processing efficiency can be simultaneously improved during the document creation automation process.

[0034] Again, this will be explained with reference to Fig. 1.

[0035] In response to a user's request, the service server (10) can obtain data that requires reference from among the data stored in the knowledge base (20) and input the obtained data to the LLM (Large Language Model) server (40) along with the user's request. The LLM server (40) may be a server operated by the same operator as the operator of the document creation system (1), or a server operated by a third party separate from the operator of the document creation system (1). The service server (10) can input information regarding encrypted chunks stored in the second database (22) to the LLM server (40) along with the user's request (e.g., prompt). By inputting information regarding encrypted chunks to the LLM server (40), the service server (10) can take measures to ensure that resources uploaded by the user are stored only locally within the document creation system (1) and are not leaked outside the document creation system (1). Accordingly, according to the present embodiment, by resolving security vulnerabilities where user documents can be transmitted to an external server and stored temporarily or permanently, the integrity and security of the data can be improved.

[0036] The service server (10) can receive a framework modification request from a user terminal (30) requesting a modification to the framework of the first document.

[0037] The service server (10) may, in response to the framework modification request, analyze the framework of the reference target resource and generate a template prompt based on the framework of the reference target resource. This will be explained in detail with reference to FIGS. 6 to 13.

[0038] The service server (10) can obtain a chunk similar to the component included in the template prompt generated as above from among the chunks related to the first document stored in the knowledge base (20). This will be explained in detail with reference to FIG. 14.

[0039] The service server (10) inputs the input chunk, the template prompt, and the user's request to modify the framework into a large-scale language model provided on the LLM server (40), and can generate draft content in which the framework of the first document is modified. In this case, the input chunk is encrypted and obfuscated when stored in the knowledge base (20), so that even if it is input into a large-scale language model provided on the LLM server (40), the leakage of data related to the first document can be prevented.

[0040] Each of the components (10, 20, 21, 22 and 23) of the above-described document creation system (1) may be implemented in at least one computing device. For example, all functions of the document creation system (1) may be implemented in a single computing device, or the first function of the document creation system (1) may be implemented in a first computing device and the second function may be implemented in a second computing device. Alternatively, specific functions of the document creation system (1) may be implemented in multiple computing devices.

[0041] A computing device may include any device equipped with computing functions, and for an example of such a device, refer to FIG. 17. Since a computing device is a collection of various components (e.g., memory, processor, etc.) that interact, it may be referred to as a 'computing system' depending on the case. Of course, the term computing system may also encompass the concept of a collection of multiple computing devices that interact.

[0042] Meanwhile, in some embodiments, the document processing system (1 to 40) can communicate through a network. Here, the network can be implemented as any type of wired or wireless network, such as a Local Area Network (LAN), a Wide Area Network (WAN), a mobile radio communication network, or Wibro (Wireless Broadband Internet).

[0043] For the sake of ease of understanding, the following description will continue under the assumption that all steps / operations of the methods described below are performed on the service server (10) described above. Therefore, if the subject of a specific step / operation is omitted, it can be understood that it is performed on the service server (10). However, in an actual environment, some steps / operations of the methods described below may be performed on other computing devices.

[0044] Hereinafter, with reference to FIGS. 3 and FIGS. 4, the UI configuration of a document creation system displayed on a user terminal according to some embodiments of the present disclosure will be described. FIGS. 3 is a diagram showing the UI configuration of a document creation system displayed on a user terminal according to some embodiments of the present disclosure. FIGS. 4 is a diagram showing an example of the UI configuration of a document creation system displayed on a user terminal according to some embodiments of the present disclosure.

[0045] Referring to FIG. 3, the user terminal screen (200) displayed on the user terminal may include a DB screen (201), a current document screen (202), an instruction screen (203), and a preview screen (204). However, it should be noted that the scope of the present disclosure is not limited thereto and is merely an example of a screen displayed on the user terminal (30).

[0046] The DB screen (201) may be a screen that displays folders, documents (e.g., Word files, PDF files, etc.) and web pages as resources uploaded by the user. Resources uploaded by the user may be stored in a specific directory in the knowledge base (20). The DB screen (201) may intuitively display directory information where specific resources are stored.

[0047] For example, referring to FIG. 4, information regarding directories such as 'Folder 1', 'Folder 2', and 'Folder 3' may be displayed on the DB screen (201). Additionally, the user can upload a folder stored on the user terminal (30) to the knowledge base (20) on the document creation system (1) by clicking the 'Add file from my computer' button displayed on the DB screen (201). The user can create a new folder or a new file by clicking the 'New folder' or 'New document' button displayed on the DB screen (201) and upload it to the knowledge base (20) on the document creation system (1).

[0048] The current document screen (202) may be a screen that displays the document currently being written by the user. The user can export the created document through the current document screen (202). The user can export the document to be edited through the current document screen (202). The user can specify the modification target for the document currently being written through the current document screen (202). The user can accept the document created by the service server (10) through the current document screen (202).

[0049] For example, referring to FIG. 4, various document tabs, document editor tools, document export buttons, and the content of a selected document may be displayed on the current document screen (202). As will be explained later, the user can perform the tasks listed above through the responsive user interface displayed on the current document screen (202).

[0050] The instruction screen (203) may be a screen where the user writes instructions so that the document work system (1) can perform tasks. The user can upload resources by dragging and dropping materials onto the instruction screen (203). The user can enter prompts regarding document creation, etc. through the instruction screen (203). The user can enter requests for modification of the document through the instruction screen (203). The user can perform tasks such as specifying materials or folders that need to be referenced for document creation, or specifying a template, through the instruction screen (203).

[0051] For example, referring to FIG. 4, the user can input requests for the document (=selected document file) displayed on the current document screen (202) on the instruction screen (203). In this case, the user can input text in the area dragged from the selected document file using 'Ctrl+E'. The user can specify materials that are essential for reference in document work by inputting resources stored in the knowledge base (20) after '@'.

[0052] The preview screen (204) may be a screen that displays information regarding the work results, etc., of a document in accordance with the user's request entered in the instruction screen (203) regarding the document displayed in the current document screen (202). The preview screen (204) may display the proposal results output by the LLM server (40) in accordance with the user's request. Additionally, the preview screen (204) may display an explanation regarding the proposal results output by the LLM server (40) in accordance with the user's request.

[0053] For example, referring to FIG. 4, the user can view various information displayed on the preview screen (204). Additionally, the user can refresh the conversation window and create a new conversation by clicking the 'New Conversation' button displayed on the preview screen (204). Additionally, the user can retrieve an existing conversation by clicking the 'Conversation History' button displayed on the preview screen (204).

[0054] Below, the explanation will continue based on the user terminal screen (200) shown in FIGS. 3 and FIGS. 4.

[0055] Hereinafter, with reference to FIG. 5, a method for modifying an instruction-based document framework according to some embodiments of the present disclosure is described. FIG. 5 is a flowchart for explaining the operation of a method for modifying an instruction-based document framework according to some embodiments of the present disclosure.

[0056] Referring to FIG. 5, a service server (10) may receive a framework modification request from a user terminal (30) requesting a modification to the framework of a first document (S3000). The framework modification request is a request to change at least one of the visual format, logical structure, or style of the document and may include an instruction identifying a reference target resource. In this case, the instruction may be identified by a designated special character included in the framework modification request. For example, the user may specify a reference target resource by entering a special character such as '@' in front of the reference target resource. However, the scope of the present disclosure is not limited thereto, and the designated special character may vary depending on the settings.

[0057] Subsequently, the service server (10) can use an AI model to analyze a reference target resource and extract elements that constitute the framework of the resource. The framework may include a visual format layout, a logical structure, and a writing style, and the service server (10) can generate a template prompt based on the framework of the reference target resource (S3100). The template prompt serves as a type of document writing guideline that can be applied when writing a document in the future and can be stored in a structured format. The AI ​​model may be a pre-trained language model based on natural language processing or a deep learning model for document analysis, and can automatically identify the visual format layout, logical structure, and writing style by recognizing the expression method, text placement, syntax patterns, and hierarchical structure of sentences included in the reference target resource.

[0058] Subsequently, the service server (10) searches for chunks among the chunks related to the first document stored in the knowledge base (20) that have a similarity level greater than or equal to a threshold value with respect to the components included in the generated template prompt, and can obtain an input chunk that has a similarity level greater than or equal to a threshold value with respect to the components included in the template prompt (S3200). The input chunk is a semantic unit element of the first document that may be affected when the framework is changed.

[0059] Subsequently, the service server (10) can generate draft content with the framework of the first document modified by inputting the input chunk, the template prompt, and the framework modification request into a pre-trained large language model (LLM) (S3300). The generated draft content can be expressed in a new format that reflects the framework of the reference resource while maintaining the same content basis as the existing document. The large language model may be a model provided by the LLM server (40). The large language model may be a natural language generation model that is pre-trained based on large text data and can perform tasks such as context understanding, sentence generation, and style reflection.

[0060] According to the present embodiment, a user can easily change the style, organizational flow, and visual layout of a document based on a reference document (reference target resource) and obtain a high-quality document draft in an automated manner. That is, according to the present embodiment, the user can easily unify the style (writing style), organizational flow (logical structure), and visual layout (format layout) of the document based on the reference document, and can automatically generate a high-quality document draft with a consistent framework based on the content of an existing document without separate manual work.

[0061] In addition, according to the present embodiment, even if the document creator does not specifically define the structure or format of the document, the system automatically learns the framework of the reference document and templates it, thereby enabling even users without specialized knowledge to easily create documents of a fixed format, and making it possible to standardize documents and maintain quality consistency.

[0062] In addition, this embodiment can provide universality and scalability that can be easily applied to various document domains (e.g., contracts, technical reports, policy documents, etc.).

[0063] In addition, according to the present embodiment, by utilizing a large-scale language model (LLM), it is possible to generate high-quality results that conform to a reference framework while maintaining the semantic flow of existing documents, thereby dramatically improving the efficiency of the document creation and editing process.

[0064] Hereinafter, with reference to FIGS. 6 and FIGS. 7, a method for generating a template prompt according to some embodiments of the present disclosure is described. FIGS. 6 is a detailed flowchart for explaining the detailed operation of a method for modifying an instruction-based document framework according to some embodiments of the present disclosure, as described with reference to FIGS. 4. FIGS. 7 is a diagram for explaining an example of a format prompt according to some embodiments of the present disclosure.

[0065] Referring to FIG. 6, the service server (10) can analyze the reference target resource in response to a framework modification request and extract format attribute values ​​that constitute the visual format layout of the resource (S3110).

[0066] Specifically, the service server (10) may obtain the following attribute values ​​for each text element (e.g., title, body, page number, etc.) included in the reference target resource. However, the scope of the present disclosure is not limited thereto, and other attribute values ​​other than those listed may also be obtained.

[0067] 1. Location information (e.g., top, center, bottom)

[0068] 2. Alignment information (e.g., left alignment, center alignment, etc.)

[0069] 3. Font attributes (e.g., size, boldness, font)

[0070] 4. Margins (e.g., top / bottom margins)

[0071] 5. Detect repeating elements (e.g., position or pattern of page numbers)

[0072] The service server (10) can generate a format prompt that reflects the visual format layout of the reference target resource based on the format attribute values ​​extracted as above (S3111). The visual format layout is generated based on the format attribute values ​​and may include, for example, information regarding the alignment position of the title (subject), the margin value of the body, the page number position, the numbering method, and the relative position between the contents.

[0073] The service server (10) can maintain a consistent visual style when reconfiguring or automatically generating the framework of the reference target resource using the format prompt.

[0074] Referring to FIG. 7, the format prompt can be expressed in the form of structured JSON. For example, the service server (10) recognizes visual attributes within the reference target resource,

[0075] 1. Title alignment is center,

[0076] 2. Title is 18pt

[0077] 3. The main text is 12pt in size,

[0078] 4. The top margin is 2.5cm,

[0079] 5. The bottom margin is 2.5cm,

[0080] 6. Page numbers are located at the bottom center, and

[0081] 7. The spacing between sections is 1.5 cubic centimeters

[0082] You can configure the format prompt based on information such as the above.

[0083] The service server (10) can change the visual layout of the first document to the visual layout of the reference target resource using the format prompt.

[0084] Hereinafter, with reference to FIGS. 8 and FIGS. 9, a method for generating a template prompt according to some embodiments of the present disclosure will be described. FIGS. 8 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, described with reference to FIGS. 4. FIGS. 9 is a diagram for explaining an example of a logical structure prompt according to some embodiments of the present disclosure.

[0085] Referring to FIG. 8, after receiving a request to modify a framework, the service server (10) can analyze the reference target resource to extract the logical structure of the reference target resource (S3120). More specifically, the service server (10) can analyze the hierarchical structure between each component item (e.g., chapter, section, subsection, etc.) within the reference target resource and the representation method (display pattern) in which the hierarchical structure is expressed.

[0086] The subject of analysis may include the following elements. However, the scope of the present disclosure is not limited thereto, and other elements may be included in the subject of analysis.

[0087] 1. Inclusion relationships between parent and child items (e.g., form 2 → 2.1 → 2.1.1)

[0088] 2. Expression patterns (e.g., numbering, indentation, bold headings, etc.)

[0089] 3. Title structure and method of separation from the body text (e.g., blank lines, line breaks, bold text, etc.)

[0090] Based on the analysis results above, the service server (10) can generate a logical structure prompt that reproduces the hierarchical configuration flow of the reference target resource (S3121).

[0091] The service server (10) can determine how the reference target resource is organized in a structural flow using the above logical structure prompt, and can maintain a consistent hierarchical structure when a large-scale language model generates draft content with the framework of the first document modified.

[0092] Referring to FIG. 9, the logical structure prompt can be configured in JSON form. For example, a service server (10) can obtain information regarding a hierarchical structure in which a referenced resource is composed of three upper sections, and among them, 'Section 2' is composed of two lower sub-sections, and can use this to modify the framework of the first document by template.

[0093] Hereinafter, a method for generating a template prompt according to some embodiments of the present disclosure is described. FIG. 10 is a detailed flowchart for explaining the detailed operation of a method for modifying an instruction-based document framework according to some embodiments of the present disclosure, described with reference to FIG. 4. FIG. 11 is a diagram for explaining an example of a style prompt according to some embodiments of the present disclosure.

[0094] Referring to FIG. 10, the service server (10) can extract style attributes for text items included in the reference target resource in response to a framework modification request (S3130).

[0095] Specifically, the above stylistic attribute value can be derived based on the following textual characteristics.

[0096] 1. Vocabulary usage patterns: technical terms, everyday language, abstract / concrete expressions, etc.

[0097] 2. Sentence Length Distribution: Average sentence length, proportion of complex sentences, number of sentences within a paragraph

[0098] 3. Sentence-ending expressions: Types of sentence-ending particles such as imperative, declarative, and interrogative forms

[0099] 4. Expression tone: Polite, neutral, persuasive, firm style, etc.

[0100] Subsequently, the service server (10) can generate a style prompt regarding the style of the reference target resource based on the style attribute value extracted as above (S3131). The style prompt is an element that influences the “speech style” or “expression style” of the document and can contribute to maintaining or changing the consistency and tone and manner of the document.

[0101] Referring to Fig. 11, the generated style prompt can be configured in JSON form. For example, "tone": "formal" means a formal expression, "sentenceStructure": "concise-declarative" means a sentence centered on short and assertive statements, and "terminology": "technical" means a writing style centered on technical terms.

[0102] The service server (10) can automatically generate a draft of a document with the framework modified by consistently applying the stylistic style of the reference resource to the content of the first document based on a style prompt regarding the reference target resource.

[0103] Referring to FIG. 12, the service server (10) can generate a single template prompt by merging a format prompt such as FIG. 7, a logical structure prompt such as FIG. 9, and a style prompt such as FIG. 11. That is, the template prompt may be a prompt containing specific information regarding the visual format layout, logical structure, and style of the reference target resource. The service server (10) inputs the template prompt configured as above into a large language model along with a chunk related to the first document and a user's request to modify the framework, and can generate draft content in which the framework of the first document is modified like the reference target resource.

[0104] According to the present embodiment, based on a template prompt generated by integrating the three elements above, the service server (10) inputs the semantic unit chunks of the first document and the framework modification request received from the user together into a large-scale language model (LLM), thereby automatically generating draft document content that reflects the framework of the reference target resource while maintaining the core content of the first document.

[0105] In addition, according to the present embodiment, compared to existing simple text replacement methods or document generation technologies centered on format reuse, the quality and consistency of the entire document can be dramatically improved by integrally reflecting heterogeneous attributes such as the format, structure, and style of the document. Furthermore, since the user can perform automated framework conversion with only a single reference resource without having to specify detailed items or styles individually, it can provide the effect of increasing both the professionalism and productivity of document creation.

[0106] In particular, according to the present embodiment, the method of modularizing and merging format, logic, and stylistic prompts can be utilized to build a template library combining various resources in the future or to easily standardize customized document styles for each organization, and can provide an input structure that enables more stable and precise utilization of the performance of a large-scale language model. Therefore, the present embodiment can provide a highly effective means to automatically generate a high-quality, unified document framework while minimizing user intervention.

[0107] Hereinafter, with reference to FIG. 13, a method for utilizing a template prompt according to some embodiments of the present disclosure will be described. FIG. 13 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, which was described with reference to FIG. 4.

[0108] Referring to FIG. 13, the service server (10) can obtain document type information for a reference target resource during the process of generating a template prompt (S3140). The document type information refers to information that identifies the type of document that the reference resource has a specific purpose and format, such as a resume, a resignation letter, a proposal, a contract, a technical report, etc. The document type information can be obtained automatically or manually based on the metadata, filename, title, content structure, or user input of the reference target resource.

[0109] Afterwards, the service server (10) stores the previously generated template prompt in a knowledge base, but can store it as a document template corresponding to the document type information (S3141). That is, instead of simply storing the format, structure, and style prompts, it stores the template prompt by explicitly tagging what kind of document it corresponds to, so that when a user makes a request such as "write a resume" or "make a resignation form," the prompt corresponding to the document type of the request can be immediately retrieved and applied.

[0110] For example, when a user uploads a resume once, the system analyzes it to extract its framework (format, structure, style) and saves it as a template with the document type "Resume." Subsequently, if the same user requests to "rewrite my resume," a large-scale language model can automatically generate a document based on the saved "Resume" template prompt.

[0111] In this way, by generating template prompts based on reference target resources and systematically storing them by document type, the document creation system (1) can generate customized automatic documents for various document types in the future, and the user has the advantage of being able to easily generate high-quality documents with just a simple request without having to specify templates one by one.

[0112] Hereinafter, with reference to FIG. 14, a method for obtaining an input chunk according to some embodiments of the present disclosure is described. FIG. 14 is a detailed flowchart for describing the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, which was described with reference to FIG. 4.

[0113] Referring to FIG. 14, the service server (10) can perform semantic embedding-based similarity analysis to obtain an input chunk corresponding to a template prompt from the first document.

[0114] That is, the service server (10) can calculate the similarity between the semantic embedding of the first document stored in the third database (23) and the components included in the generated template prompt (S3210). The components may correspond to the keys in the structured template prompt. That is, the service server (10) can calculate the similarity between the semantic embedding of each chunk included in the first document and the values ​​corresponding to the keys included in the template prompt.

[0115] For example, if a template prompt includes components such as “System Overview”, “Operation Principle”, and “Key Advantages”, the service server (10) can vectorize the expressions corresponding to each component and calculate a similarity by comparing them with chunk embeddings of the first document in the third database (23). The similarity is calculated based on, for example, cosine similarity, and if the value is 0.85 or higher, it can be determined that they are semantically similar. However, the scope of the present disclosure is not limited to this, and the criteria for determining similarity may be set differently depending on the situation.

[0116] The service server (10) can obtain a chunk of the first document corresponding to a semantic embedding determined to be greater than or equal to a threshold value as an input chunk based on the above similarity analysis result (S3211). For example, if a template prompt contains "2. Operation Principle" and the similarity with the embedding of the chunk "3. Operation Principle and Control Logic" of the first document is greater than or equal to a threshold value, the corresponding chunk can be extracted and inserted into the template prompt.

[0117] This method does not simply compare keywords or titles, but rather compares based on the semantic content within the paragraphs; therefore, semantically similar content can be effectively extracted even if the expression styles within the document differ.

[0118] The service server (10) can use the above method to automatically reconstruct the existing content of the first document to fit the framework of the reference target resource, and as a result, the format, structure, and style of the reference target resource are maintained as they are, while a draft document reflecting the actual content of the first document is automatically generated.

[0119] Accordingly, according to the present embodiment, by accurately identifying existing chunks similar to prompt components through semantic embedding-based similarity analysis, the framework can provide a technical effect of generating a newly configured high-quality document while preserving the context and semantics of the document content.

[0120] Hereinafter, with reference to FIG. 15, a method for generating draft content according to some embodiments of the present disclosure will be described. FIG. 15 is a detailed flowchart for explaining the detailed operation of an instruction-based document framework modification method according to some embodiments of the present disclosure, which was described with reference to FIG. 4.

[0121] In some cases, it may be necessary to determine whether to reflect the modified draft content of the first document, which was generated using a large-scale language model (LLM), through a user confirmation and acceptance process before finally reflecting it in the first document.

[0122] Referring to FIG. 15, the service server (10) can transmit a preview to the user terminal (30) in which content corresponding to the input chunk is reflected in the components of the template prompt. That is, the service server (10) can generate a preview in which parts of the framework that have been changed are reflected among the draft content generated based on the first document, the template prompt, and the input chunk, and transmit it to the user terminal. The preview can be displayed on a preview screen (204) on the user terminal screen (200). The preview may be a part or the whole of a document in which a newly applied visual format, logical structure, or writing style is reflected according to the template prompt.

[0123] After reviewing the above preview, the user may explicitly respond with an intention to accept or reject, and the service server (10) may receive a response from the user terminal regarding whether to accept or reject (S3311).

[0124] Afterwards, the service server (10) can determine whether to accept the user's preview (S3313).

[0125] If the user gives an acceptance response (S3314), the service server (10) can finally reflect the content included in the preview in the first document. In other words, the generated draft content can be substantially applied to the composition of the actual document. In this case, the first document reflecting the content included in the preview can be displayed on the current document screen (202) on the user terminal screen (200).

[0126] On the other hand, if the user does not accept the modified draft content displayed in the preview (S3315), the draft content is not applied, and the first document may be maintained without change.

[0127] According to the present embodiment, a complementary process is provided that allows for the selective incorporation of revisions proposed by AI while maintaining the user's decision-making authority even during an automated document modification process.

[0128] In particular, for important official documents, contracts, and policy documents, designing the system so that automatic revisions are reflected only after the user personally reviews and accepts them has the effect of simultaneously ensuring the reliability and quality control of automated document generation.

[0129] Hereinafter, with reference to FIG. 16, a method for generating draft content according to some embodiments of the present disclosure will be described.

[0130] In one embodiment, the service server (10) may display a draft content (111) in which the framework of the first document has been modified on the current document screen (202) on the user terminal screen (200). At this time, the service server (10) may display various information on the preview screen (204), such as the progress of the creation of the modified content (111), the real-time creation process, and information regarding the framework content modified in the first document. The framework content modified in the first document may, for example, be regarding the format, style, logical structure, etc. modified in the first document.

[0131] According to the present embodiment, the user can clearly verify the location and process of the generation process, thereby increasing the reliability of the generation results and allowing for immediate requests to stop or modify the process if necessary. Therefore, according to the present embodiment, there is an effect of improving the transparency, predictability, and convenience in terms of user interaction of automatic document generation.

[0132] Hereinafter, with reference to FIG. 17, an exemplary computing device (1000) capable of implementing a system according to some embodiments of the present disclosure will be described.

[0133] FIG. 17 illustrates an exemplary computing device capable of implementing systems according to some embodiments of the present disclosure. The computing device (1000) of FIG. 17 may include one or more processors (1100), a system bus (1600), a communication interface (1200), a memory (1400) for loading a computer program (1500) executed by the processor (1100), and a storage (1300) for storing the computer program (1500). For example, the computing device of FIG. 17 may be the service server (10) described with reference to FIG. 1.

[0134] The processor (1100) controls the overall operation of each component of the computing device (1000). The processor (1100) may perform operations on at least one application or program for executing the method / operation according to various embodiments of the present disclosure. The memory (1400) stores various data, instructions and / or information. The memory (1400) may load one or more computer programs (1500) from the storage (1300) to execute the method / operation according to various embodiments of the present disclosure. The storage (1300) may store one or more computer programs (1500) non-temporarily. The computer program (1500) may include one or more instructions in which the method / operation according to various embodiments of the present disclosure is implemented. When a computer program (1500) is loaded into memory (1400), the processor (1100) can perform methods / operations according to various embodiments of the present disclosure by executing one or more of the instructions.

[0135] The computer program (1500) may be a program for modifying an instruction-based document framework.

[0136] In some embodiments, the computer program (1500) may include instructions for performing operations such as receiving a framework modification request from a user terminal requesting modification to the framework of a first document—the framework modification request including instructions identifying a reference target resource—analyzing the framework of the reference target resource using an AI model and generating a template prompt based on the framework of the reference target resource, searching for a chunk among chunks related to the first document stored in a knowledge base that has a similarity to a component included in the template prompt greater than or equal to a threshold value and obtaining an input chunk, and inputting the input chunk, the template prompt, and the framework modification request into a pre-trained Large Language Model and generating a draft content in which the framework of the first document is modified.

[0137] In some embodiments, the computing system described with reference to FIG. 17 may be configured using one or more physical servers included in a server farm based on cloud technology, such as a virtual machine. In this case, at least some of the components shown in FIG. 17, such as the processor (1100), memory (1400), and storage (1300), may be virtual hardware, and the communication interface (1200) may also be configured as a virtualized networking element, such as a virtual switch.

[0138] Although the operations are depicted in a specific order in the drawings, it should not be understood that the operations must be executed in the specific order depicted or in sequential order, or that all depicted operations must be executed to obtain the desired result. In certain situations, multitasking and parallel processing may be advantageous.

[0139] Although embodiments of the present disclosure have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be practiced in other specific forms without altering the technical concept or essential features thereof. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of protection of the present invention shall be interpreted by the claims below, and all technical concepts within the equivalent scope shall be interpreted as being included within the scope of rights of the technical concepts defined by the present disclosure.

Claims

Claim 1 A method performed by a computing system comprises: receiving a framework modification request from a user terminal requesting a modification to the framework of a first document—the framework modification request includes instructions identifying a reference target resource—; analyzing the framework of the reference target resource using an AI model and generating a template prompt based on the framework of the reference target resource; searching for a chunk among chunks related to the first document stored in a knowledge base that has a similarity to a component included in the template prompt greater than or equal to a threshold value and obtaining an input chunk; and automatically relocating the input chunk to a corresponding position on the framework of the first document according to format rules, hierarchical rules, and stylistic rules included in the template prompt. The method comprises the steps of inputting the relocated input chunk, the template prompt, and the framework modification request into a pre-trained Large Language Model, and generating a draft content in which the framework of the first document is modified, wherein the step of generating the template prompt includes: extracting format attribute values ​​regarding location information, alignment information, font attributes, margins, and repeating elements included in the reference target resource, and generating a format prompt reflecting the visual format layout of the reference target resource based on the format attribute values; analyzing the hierarchical structure between constituent items included in the reference target resource and the representation method of the hierarchical structure, and generating a logical structure prompt regarding the logical structure of the reference target resource based on the hierarchical structure and the representation method; and extracting stylistic attribute values ​​regarding vocabulary usage patterns, sentence length distribution, sentence ending expressions, and expression tone regarding text items included in the reference target resource, and generating a style prompt regarding the style of the reference target resource based on the stylistic attribute values.A method for modifying an instruction-based document framework, comprising the step of integrating the format prompt, the logical structure prompt, and the style prompt to generate a template prompt that reflects the format, structure, and style of the referenced target resource. Claim 2 An instruction-based document framework modification method according to claim 1, wherein the instruction is identified by a designated special character included in the framework modification request. Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 A method for modifying an instruction-based document framework according to claim 1, wherein the step of generating the template prompt comprises: obtaining document type information of the reference target resource; and storing the template prompt in the knowledge base as a template corresponding to the document type information. Claim 7 In claim 1, the knowledge base comprises: a first database in which upload resources are parsed by semantic units according to predefined criteria and chunks corresponding to each semantic unit are stored; a second database in which directory information in which the chunk is stored in the knowledge base is mapped and stored for each of the chunks; and a third database in which semantic embeddings vectorized from the chunks are stored. The step of obtaining the input chunk comprises: a step of calculating the similarity between the semantic embedding for the first document stored in the third database and the components included in the template prompt; and a step of obtaining the chunk corresponding to the semantic embedding in which the similarity is greater than or equal to a threshold value from the first document. Claim 8 A method for modifying an instruction-based document framework according to claim 1, wherein the step of generating a draft content modified by the framework of the first document comprises: transmitting a preview to the user terminal in which content corresponding to the input chunk is reflected in the template prompt; receiving a response from the user terminal regarding whether to accept the preview; and, if the response is an instruction accepting the preview, reflecting the content included in the preview into the first document. Claim 9 An instruction-based document framework modification method according to claim 1, wherein the step of generating a draft content modified by the framework of the first document includes the step of displaying the generation progress of the draft content and the framework content modified in the first document in real time. Claim 10 A communication interface; a memory on which a computer program is loaded; and one or more processors on which the computer program is executed, wherein the computer program comprises: an operation of receiving a framework modification request from a user terminal requesting a modification to the framework of a first document—the framework modification request includes an instruction identifying a reference target resource—; an operation of analyzing the framework of the reference target resource using an AI model and generating a template prompt based on the framework of the reference target resource; an operation of searching for a chunk among chunks related to the first document stored in a knowledge base that has a similarity to a component included in the template prompt greater than or equal to a threshold value and obtaining an input chunk; and a step of automatically relocating the input chunk to a corresponding position on the framework of the first document according to format rules, hierarchical rules, and stylistic rules included in the template prompt. The method includes instructions for inputting the relocated input chunk, the template prompt, and the framework modification request into a pre-trained Large Language Model, and for performing an operation to generate a draft content in which the framework of the first document is modified, wherein the operation to generate the template prompt includes: an operation to extract format attribute values ​​regarding location information, alignment information, font attributes, margins, and repeating elements included in the reference target resource, and to generate a format prompt reflecting the visual format layout of the reference target resource based on the format attribute values; an operation to analyze the hierarchical structure between constituent items included in the reference target resource and the representation method of the hierarchical structure, and to generate a logical structure prompt regarding the logical structure of the reference target resource based on the hierarchical structure and the representation method.An instruction-based document framework modification system comprising: an operation to extract stylistic attribute values ​​regarding vocabulary usage patterns, sentence length distribution, sentence ending expressions, and expression tone with respect to text items included in the reference target resource, and to generate a style prompt regarding the style of the reference target resource based on the stylistic attribute values; and an operation to integrate the format prompt, the logical structure prompt, and the style prompt to generate a template prompt reflecting the format, structure, and style of the reference target resource.

Images