Natural language-based Excel automation task support apparatus

The natural language-based Excel automation task support device addresses the inefficiencies of conventional Excel systems by providing a structured UI for automating tasks through AI, ensuring reliable execution and user control, and enabling advanced tasks for non-developers.

KR103044779B1Active Publication Date: 2026-11-10SPINX CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
KR1020250153729
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2025-10-22
Publication Date
2026-11-10
Estimated Expiration
2045-10-22

AI Technical Summary

Technical Problem

Conventional Excel systems lack a structured user interface and execution flow management system to accurately interpret user natural language requests and automate tasks such as formula generation and data processing, relying heavily on VBA or external add-ins, which is inefficient and user proficiency-dependent.

Method used

A natural language-based Excel automation task support device that includes a user interaction interface, request analysis and classification module, AI execution plan generation module, and task execution flow control module to analyze and execute user commands, providing a structured UI for task automation and error handling.

Benefits of technology

Ensures reliable and predictable automation execution by visually representing the automation flow, allowing user control over the process, and structuring execution plans for reuse and failure recovery, making advanced Excel tasks accessible to non-developers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112025117924656-PAT00001_ABST
    Figure 112025117924656-PAT00001_ABST
Patent Text Reader

Abstract

A natural language-based Excel automation task support device is disclosed. The disclosed device is a natural language-based Excel automation task support device operating in a Microsoft Excel environment, comprising: a user interaction interface module that receives a natural language command from a user, detects a reference object selected upon input of the natural language command, converts the detected reference object into an internal reference structure, and outputs it as user request data; a request analysis and classification module that receives the user request data, analyzes the intent, behavior, and context of the natural language command, converts the analysis result into structured command interpretation information, determines a corresponding execution type based on the analysis result, and outputs it as an AI execution judgment result; an AI execution plan generation module that receives the AI ​​execution judgment result, concretizes the natural language command into actual execution procedure units, and generates an AI execution schema; and a task execution flow control module that receives the AI ​​execution schema, selects a task execution target object, calls an internal execution routine corresponding to the determined execution type to perform the actual task, and outputs the actual task result as task execution result data. It includes an AI response processing module that determines whether a natural language command is an informational query based on the AI ​​execution judgment result of the request analysis and classification module, and if the natural language command is an informational query, generates and outputs a natural language response sentence.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a natural language-based Excel automation task support device. Background Technology

[0002] In the conventional Microsoft Excel environment, users faced the inconvenience of having to manually configure various advanced functions, such as formula creation, data filtering, and conditional formatting, when performing data analysis and document creation tasks. The accuracy and efficiency of these tasks varied significantly depending on the user's proficiency, and automating repetitive or complex tasks primarily required reliance on VBA (Visual Basic for Applications) or separate external add-ins.

[0003] Meanwhile, due to the recent advancement of AI technologies based on natural language processing—such as conversational AI models like ChatGPT—there are active attempts to automate various software tasks based on user natural language commands. This trend is also evident in the Excel environment, where there is a growing demand to automate tasks such as formula generation, data processing, and condition specification through natural language-based commands.

[0004] However, conventional Excel systems and some existing AI integration solutions currently lack a structured user interface (UI) and execution flow management system capable of accurately interpreting users' natural language requests within actual business contexts and automatically converting those requests into specific execution flows or operation procedures. In particular, there is a limitation in that there is a lack of a systematic UI / UX structure designed to integrate AI functions in a way that is easy for general users to understand and to operate naturally without conflicting with Excel's basic operations.

[0005] The matters described in the background technology above are intended to aid in understanding the background of the invention and may include matters that are not disclosed prior art. Prior art literature

[0006] Prior Art 1: Korean Registered Patent No. 10-2838194 (Device for determining user question intent for interactive search and intelligent search) Prior Art 2: Korean Published Patent No. 10-2025-0075475 (System and method for generating a domain-based user-customized database based on question and intent classification) The problem to be solved

[0007] The present invention is proposed in consideration of the aforementioned conventional circumstances, and aims to provide a natural language-based Excel automation task support device that analyzes a user's natural language input in a Microsoft Excel environment and supports the performance of automation tasks such as data analysis, automatic formula generation, and VBA execution by utilizing artificial intelligence (AI). means of solving the problem

[0008] To achieve the above objectives, a natural language-based Excel automation task support device according to a preferred embodiment of the present invention is a natural language-based Excel automation task support device operating in a Microsoft Excel environment, comprising: a user interaction interface module that receives a natural language command from a user, detects a reference object selected upon input of the natural language command, converts the detected reference object into an internal reference structure, and outputs it as user request data; a request analysis and classification module that receives the user request data, analyzes the intent, action, and context of the natural language command, converts the analysis result into structured command interpretation information, determines a corresponding execution type based on the analysis result, and outputs it as an AI execution judgment result; an AI execution plan generation module that receives the AI ​​execution judgment result and generates an AI execution schema by concretizing the natural language command into actual execution procedure units; and a task execution flow control module that receives the AI ​​execution schema, selects a task execution target object, calls an internal execution routine corresponding to the determined execution type to perform the actual task, and outputs the actual task result as task execution result data. and an AI response processing module that determines whether the natural language command is an informational query based on the AI ​​execution judgment result of the request analysis and classification module, and if the natural language command is an informational query, generates and outputs a natural language response sentence.

[0009] The above request analysis and classification module analyzes the natural language command to identify the intent and reference target of the natural language command, and based on the identified results, classifies the natural language command into one of the types of information requests, conversation requests, and work requests.

[0010] If the above natural language command is classified as a conversation request, the AI ​​response processing module may perform a termination step after the chat window response.

[0011] If the above natural language command is classified as an information request, the AI ​​response processing module may construct a search path that matches the intent and purpose of the request, establish a search plan including the expected search purpose and search method, result output format, and related data structuring method, and display the established search plan on the UI through the user interaction interface module.

[0012] When the user interaction interface module receives a confirmation event of a search plan displayed on the UI, and receives a selected output method, the content of a user request, and reference data, the user interaction interface module transmits the selected output method, the content of a user request, and the reference data to the request analysis and classification module, the request analysis and classification module identifies the type of the user request, the user interaction interface module structures the reference data and transmits it to the request analysis and classification module, and the AI ​​response processing module performs information retrieval, refinement, and analysis based on the user request and reference data to generate summarized information descriptions and output components and provide them to the user, and as the user interaction interface module receives the user's execution selection event, the AI ​​execution plan generation module generates an AI execution schema based on the content of the user request, reference data, and the selected output format, and the task execution flow control module can perform an output operation within an Excel environment based on the AI ​​execution schema.

[0013] When the above natural language command is classified as a task request, the task execution flow control module determines whether the task can be processed based on Excel formulas or based on VBA automation, and displays this on the UI through the user interaction interface module; the AI ​​execution plan generation module receives information including the processing method selected and entered by the user, the output format, the natural language content of the user request, and reference data; the AI ​​execution plan generation module analyzes the received information to generate an AI execution plan and displays it on the UI through the user interaction interface module; and, depending on the user's execution selection, the AI ​​execution plan generation module generates an AI execution schema based on the AI ​​execution plan, and the task execution flow control module can execute the task based on the AI ​​execution schema.

[0014] The above AI execution judgment result may include the intent of the above natural language command, the action of the above natural language command, the context, and the execution type.

[0015] It may additionally include an exception and flow restoration processing module that detects errors or exceptions occurring during the execution of the above task execution flow control module, classifies the error type and saves it as a log, determines whether restoration is possible, and performs re-execution or restoration according to a user request.

[0016] It may additionally include a system integration management module that manages result data, conversation logs, and execution logs from the above AI response processing module, stores execution logs on an external server, and updates conversation records and user-specific execution history. Effects of the invention

[0017] According to the present invention with this configuration, reliability and predictability of automation execution can be ensured by structuring an Excel automation flow based on a user's natural language request and visually representing it. In other words, a natural language request entered by a user (e.g., "Organize sales data by month and create a chart") is displayed on a TaskPane UI, and the request is analyzed by AI to be structured into a specific Excel workflow (e.g., a flow of selecting a range, creating a pivot table, and creating a chart). Since this flow is expressed step-by-step through a visual interface (UI), the user can clearly predict how the automation will be executed.

[0018] Furthermore, by reflecting selection information of Excel cells or objects in AI requests, precise analysis and execution that understand the user's current task context become possible. In other words, when a user inputs a natural language request while a specific cell range, chart, or shape is selected, that selection information (such as cell references) is included as input for AI analysis and utilized to understand the context of the task. For example, if a user selects a specific table and says, "Show me the top 5 items here as a graph," the AI ​​provides analysis results that take the user's selection into account. This enables advanced analysis that goes beyond simple command interpretation to accurately reflect the user's intent and current task location.

[0019] Furthermore, by providing the ability to review, cancel, or modify each step before and after an AI execution request, user control over the entire workflow can be enhanced. In other words, the execution plan proposed by the AI ​​is not executed immediately but is first displayed in the execution plan area for the user to review. Users have the opportunity to review, edit, or cancel each step, and finally confirm the execution via the AI ​​execution request button. By allowing user intervention and control over the automation process, these steps prevent incorrect execution or errors and provide stable results.

[0020] Furthermore, by structuring automation execution plans into AI execution schemas, a foundation can be provided for various future uses, such as reuse, log management, and failure recovery. In other words, once an execution is finalized, the corresponding execution plan is automatically generated in the form of AIExecData. This data is designed to allow for the reuse of the same task, the storage of execution history as logs, and recovery (re-execution) in the event of failure during automation. Essentially, it serves as a foundation for building a sustainable and scalable automation ecosystem, rather than one-off automation.

[0021] Ultimately, according to the present invention, the AI ​​does not simply execute commands in Excel, but accurately understands the user's request, transparently displays the process, allows execution under the user's review and control, and structures the execution information for reuse. Brief explanation of the drawing

[0022] FIG. 1 is a configuration diagram of a system employing a natural language-based Excel automation task support device according to an embodiment of the present invention. Figure 2 is an internal configuration diagram of the natural language-based Excel automation task support device shown in Figure 1. FIG. 3 is a flowchart of natural language-based AI automation execution according to an embodiment of the present invention. FIG. 4 is a flowchart of an Excel cell or object reference selection branch according to an embodiment of the present invention. FIG. 5 is a diagram showing the process of branching the utterance history, branching the work request execution method, verifying the execution plan, and branching the execution according to an embodiment of the present invention. Figure 6 is a diagram showing the detailed processing flow for the information request step illustrated in Figure 5. Figure 7 is a diagram showing the detailed processing flow for the work request step illustrated in Figure 5. FIG. 8 is a diagram showing the process of execution flow control and AI execution schema generation according to an embodiment of the present invention. Specific details for implementing the invention

[0023] The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail.

[0024] However, this is not intended to limit the invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. The embodiments are provided to more fully explain the invention to those skilled in the art, and the following embodiments may be modified in various different forms, and the scope of the invention is not limited to the following embodiments. Rather, these embodiments are provided to make the disclosure more faithful and complete and to fully convey the spirit of the invention.

[0025] The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence 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.

[0026] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0027] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In order to facilitate an overall understanding of the present invention, the same reference numerals are used for identical components in the drawings, and redundant descriptions of identical components are omitted.

[0028] FIG. 1 is a configuration diagram of a system employing a natural language-based Excel automation task support device according to an embodiment of the present invention.

[0029] A natural language-based Excel automation task support device (10) according to an embodiment of the present invention can operate in an add-in environment of Microsoft Excel.

[0030] The natural language-based Excel automation support device (10) is connected to an external server (e.g., a cloud server or a DB server) (20) to enable communication. For example, when a user specifies data within an Excel document or sheet in an Add-in environment of Microsoft Excel, the natural language-based Excel automation support device (10) can automatically register the document and data to an external server such as the server (20).

[0031] A natural language-based Excel automation task support device (10) receives a user's natural language-based command input in a Microsoft Excel environment, analyzes and interprets the command based on artificial intelligence (AI), visualizes the result through an Excel user interface (TaskPane), generates an execution plan such as formula generation, data processing, and VBA-based automation, and can automate a series of flows that are automatically executed after a user confirmation process.

[0032] At this point, the user can input natural language requests consisting of text within the Excel document, and these requests are received through the input window in the TaskPane UI. The received natural language requests are passed to an AI analysis engine, where the user's intent and context are interpreted, and the interpreted results are proposed in the form of an actionable execution plan along with the classification of the task (e.g., information retrieval, formula application, iterative processing, macro execution, etc.).

[0033] In particular, the natural language-based Excel automation task support device (10) can collect object information (hereinafter referred to as "cell reference information") such as cell ranges, charts, shapes, and pictures selected by the user within Excel and reflect this in the contextual interpretation of natural language requests. As a result, more accurate AI analysis can be performed based on the association between the user's command and the currently selected data or visual elements. The cell reference information configured in this way is structured in JSON format or the like and provided to the AI ​​analysis module, through which more suitable formulas, data processing logic, or VBA code are derived.

[0034] The results of the AI ​​analysis are displayed in the "Execution Plan Area" within Excel's side panel (TaskPane), which is designed to allow users to pre-review the execution plans proposed by the AI. After reviewing the execution plan (e.g., specific formulas, data transformation methods, iterative processing commands, etc.), users can initiate the execution flow by executing the AI ​​execution request command. The subsequent flow can be structured in the order of the execution waiting state, execution confirmation phase, and AIExecData generation phase. In the execution waiting state, execution is not performed immediately even if a user requests it; instead, the execution plan is internally checked and configured first. The execution confirmation phase checks the execution environment, such as sheet protection status, cell lock status, and macro settings, to determine feasibility; if no issues arise, the execution is confirmed. In the AIExecData generation phase, the finalized execution plan is automatically generated in the form of a structured execution schema called AIExecData, which includes information such as formula strings, target cells, macro code, and execution options.

[0035] Through this series of flows, the natural language-based Excel automation task support device (10) according to an embodiment of the present invention can provide an integrated processing structure that can efficiently and reliably perform complex Excel automation tasks from the user's intuitive natural language input. This provides a technical effect that allows even non-developer users to easily perform advanced Excel automation tasks based on AI.

[0036] In FIG. 1, the server (20) can collect, analyze, and store AI execution logs, conversation history, and execution schemas in conjunction with the natural language-based Excel automation task support device (10), and can retransmit feedback data to the natural language-based Excel automation task support device (10). In other words, the server (20) communicates with the natural language-based Excel automation task support device (10), centrally manages logs, conversation history, and execution schemas related to AI execution, and provides a configuration that can improve the quality of AI automation through the analysis and feedback of the data.

[0037] Although not shown in the drawing, for example, the server (20) may include an execution log storage module, a conversation history management module, an AI execution schema management module, an analysis and feedback module, and a security and authentication management module.

[0038] The execution log storage module collects information such as the result, execution time, processing time, success / failure status, and error code of each AI execution performed by the natural language-based Excel automation task support device (10) and stores it in a log database. This ensures the reliability and traceability of the AI ​​automation execution.

[0039] The conversation history management module stores and manages natural language requests and AI response histories entered by users on a session basis, and enables the retrieval of conversation history by specific user account or document. The stored conversation history can be utilized for AI model improvement, context analysis of subsequent commands, and personalized suggestions for users.

[0040] The AI ​​execution schema management module can receive and store an AI execution schema (AIExecData; also called AI execution data) generated by a natural language-based Excel automation task support device (10). For example, the AI ​​execution schema may include user request content, reference data range, execution path options, result reflection location, verification procedure information, etc. The server (20) can reproduce or analyze the AI ​​operation history and execution process based on this AI execution schema.

[0041] The analysis and feedback module analyzes logs, conversation history, and AI execution schema data to evaluate the accuracy of AI responses, execution success rates, processing time, etc., and can transmit improvement feedback data to a natural language-based Excel automation task support device (10). This allows for continuous training and performance optimization of the AI ​​model.

[0042] The security and authentication management module performs user account authentication, access control, and encrypted data transmission and reception, and guarantees data integrity and privacy. If necessary, OAuth, JWT-based token authentication, or SSL / TLS secure communication protocols may be applied.

[0043] In other words, the server (20) is a server that centrally manages the execution history, conversation history, execution schema, error and feedback information of the AI-based automation process performed by the natural language-based Excel automation task support device (10), thereby improving the quality of AI automation and ensuring the traceability and security of the data.

[0044] The operation flow of the aforementioned server (20) can be as follows.

[0045] First, the server (20) may perform a step of receiving execution request data. The server (20) receives an execution request packet containing execution log data, an AI execution schema, conversation session information, etc. from a natural language-based Excel automation task support device (10). For example, the received packet may include user identification information (User ID, document ID, etc.), an AI execution schema (request content, reference cell / object information, output location, execution path options, etc.), execution results (success, failure, error code, etc.), and metadata such as execution time, processing time, and Excel version. If necessary, in the step of receiving execution request data, the server (20) may check for corruption of the transmitted data using a cryptographic hash (MD5, SHA-256, etc.) to verify the integrity of the received data.

[0046] Next, the server (20) can perform a log data storage and indexing step. The server (20) stores items such as execution results, error information, and execution time among the received execution data in an execution log storage module. The log data is indexed based on composite keys such as document ID, user ID, and request time, and is configured to facilitate subsequent searching, statistics, and analysis. The stored logs can be utilized for AI model performance analysis, execution history lookup, audit records, etc. During the log data storage and indexing step, the server (20) can improve storage efficiency by performing log compression (e.g., JSON compression, Parquet conversion, etc.) at regular intervals.

[0047] Next, the server (20) can perform a conversation history management step. The conversation history management module of the server (20) stores the user's natural language request and the AI's response data in session units. Each conversation history may record the time of utterance, the type of request (information request, task request, conversation request), a summary of the AI ​​response, and whether it was executed. Conversation history previously performed by the same user can be fed back to the AI ​​analysis / response processing module of the natural language-based Excel automation task support device (10) and utilized for contextual interpretation. Additionally, the server (20) can anonymize conversation logs stored for a certain period of time and use them as a dataset for model improvement.

[0048] Next, the server (20) can perform an AI execution schema management step. The AI ​​execution schema management module of the server (20) stores and manages AI execution schemas transmitted from the natural language-based Excel automation task support device (10). Here, the AI ​​execution schema is a set of execution information structured from a user's natural language request, and may include the request content, reference range, processing method (formula / VBA / data output), pre-execution verification history, and result reflection location. The server (20) can reproduce past AI executions based on the same schema or analyze the execution patterns of a specific user or document. In the AI ​​execution schema management step, the server (20) can version control the AI ​​execution schemas and assign a hash ID to each schema to ensure the same conditions are guaranteed when re-executing in the future.

[0049] Next, the server (20) can perform an analysis and feedback step. The analysis and feedback module of the server (20) integrates and analyzes the execution logs, conversation history, and AI execution schema accumulated in the server (20) to generate statistical feedback for improving the quality of AI automation. For example, it can automatically classify request types with high execution failure rates, analyze the causes of errors (e.g., Excel sheet locking, macro disabling, etc.), and quantify AI response accuracy or processing speed to generate a performance report. The analysis results are fed back to the AI ​​analysis / response processing module or system integration management module of the natural language-based Excel automation support device (10) and are used to optimize the processing path of the same request in the future. Additionally, the analysis and feedback module of the server (20) may configure an AI retraining dataset for specific error types and provide it to a learning system inside the server for AI model improvement.

[0050] Next, the server (20) can perform security and authentication processing steps. The security and authentication management module of the server (20) performs user authentication, access control, and data encryption and decryption. The natural language-based Excel automation task support device (10) performs OAuth or JWT token-based authentication when connecting to the server (20), and all data is transmitted through an SSL / TLS-based encryption channel. The server (20) restricts access, such as log viewing, execution schema lookup, and data recovery, according to user-specific access control (Role-based Access Control). Additionally, a data lifecycle policy can be applied so that log and conversation data are automatically deleted after a specified retention period.

[0051] Finally, the server (20) can perform data restoration and history lookup steps. If a user wishes to restore previous execution results or conversation content in an Excel environment, the server (20) reconstructs and provides the execution context at that point in time based on the stored logs and execution schema. This function is useful for restoring AI execution results in the event of Excel document corruption or local failure. The restored data reproduces the cell values, formulas, VBA code, and output location information exactly as they were at the time of the request.

[0052] The server (20), which is executed in the steps described above, can have the effect of ensuring systematic management and verification of the AI ​​automation history. That is, all AI execution history and conversation history remain as logs on the server, making auditing and quality verification possible.

[0053] Furthermore, it can lead to the continuous improvement of AI response quality. In other words, by utilizing failure logs and error patterns as training data, the accuracy and stability of AI automation are enhanced.

[0054] Furthermore, it can provide the effect of offering user-customized AI responses. In other words, by analyzing the user's recurring patterns based on conversation history, personalized recommendations can be provided for future queries.

[0055] In addition, it offers the benefit of providing data restoration and backup functions. In other words, by utilizing server save history, it prevents data loss in the Excel environment and enables stable recovery.

[0056] Furthermore, it can have the effect of enhancing security and maintaining integrity. In other words, it prevents the leakage and tampering of sensitive user data through encrypted communication and access control.

[0057] Figure 2 is an internal configuration diagram of the natural language-based Excel automation task support device shown in Figure 1.

[0058] A natural language-based Excel automation task support device (10) may include a user interaction interface module (30), a request analysis and classification module (40), an AI execution plan generation module (50), a task execution flow control module (60), an exception and flow restoration processing module (70), an AI response processing module (80), and a system integration management module (90).

[0059] The user interaction interface module (30) receives a natural language command from the user, detects a reference object selected when the natural language command is entered, and converts the detected reference object into an internal reference structure and outputs it as user request data. Here, the user request data may include command information (a natural language command entered by the user (e.g., "Summarize this table", "Show me only this month's sales as a graph"), input method (indicating whether it is text input or voice input (e.g., "text" / "voice")), reference object information (a list including identifiers (ID), locations (cell addresses), and object types (Type) of cells, tables, shapes, charts, etc. selected by the user), reference object structure (JSON structure data generated by a reference history conversion module (e.g., cell selection history JSON conversion module) - expressing the structural relationships of the selected objects), user metadata (user ID, current sheet name, active workbook, language settings, permission information, etc.), session and time information (session to which the request belongs and time of the request), etc.

[0060] In other words, the user interaction interface module (30) can perform the role of a user interface that receives natural language commands or voice input from the user within the Microsoft Excel environment, detects objects such as cells, charts, and shapes selected by the user in real time, and converts the selection information into structured reference data (JSON, etc.).

[0061] That is, the user interaction interface module (30) supports the user in entering natural language-based or voice-based commands within the Excel environment, receives the user's request, and transmits the corresponding input data to the AI ​​analysis module (42) within the request analysis and classification module (40).

[0062] Additionally, the user interaction interface module (30) detects the information in real time when the user selects or references objects such as cells, charts, and shapes within Excel, and structures and transmits the selection information so that it can be utilized in the subsequent processing flow.

[0063] In addition, the user interaction interface module (30) enables integrated user interaction with various input methods such as voice recognition, natural language input, and object selection detection, and provides an interface environment that increases user convenience.

[0064] The operation flow of the user interaction interface module (30) may be as follows.

[0065] First, the command reception step can be performed. In the command reception step, the user can directly input natural language text into the Excel UI or speak commands aloud. Voice input is converted into text by Speech-to-Text (STT) processing.

[0066] Next, an object selection detection step can be performed. An action occurs in which a user clicks on an object, such as a specific cell, chart, or picture, or selects it with a mouse. Accordingly, the user interaction interface module (30) detects the selection event and determines the type (cell, chart, image, etc.), location (address or area), and attributes (format, value, formula, etc.) of the selected object in real time.

[0067] Next, a reference data structuring step can be performed. The user interaction interface module (30) converts the detected object information into a normalized JSON format.

[0068] Finally, the input information integration and transmission step can be performed. The user interaction interface module (30) combines the received user input command (e.g., "Calculate the total of this cell") and structured reference object information into a single request packet or command object. This integrated request is transmitted to the request analysis and classification module (40) and used for subsequent processing.

[0069] Of course, although optional, the user interaction interface module (30) can provide user feedback. For example, the user interaction interface module (30) can provide visual feedback (highlighting or pop-up on the UI) regarding whether a command has been received, whether a reference object has been selected, etc. Additionally, it can display an appropriate guidance message (e.g., "Please select a cell") when an error occurs.

[0070] The above-described user interaction interface module (30) may include a user interface module (31), a cell reference selection detection module (32), and a reference history conversion module (33).

[0071] The user interface module (31) is a front-end interface responsible for the interaction between the user and the artificial intelligence (AI) analysis system within the Microsoft Excel environment, and allows the user to input natural language-based requests related to objects such as cells, shapes, and charts specified on the Excel sheet and visually check the results. That is, the user interface module (31) can receive input from the user and transmit cell reference information, selected object information, command type, natural language request statement, etc. to the AI ​​execution plan generation module (50).

[0072] For example, the user interface module (31) can be implemented as an Excel TaskPane UI, a Custom Tab, or an independent dialog box.

[0073] The main functions of the above-described user interface module (31) may include a natural language input reception function, a selection object detection and reference information collection function, a command structuring function, a processing request transmission function, a result reception and feedback display function, etc.

[0074] In the case of the natural language input reception function, the user inputs a command (e.g., "Calculate the average of this range") in natural language through the input window of the TaskPane or Custom tab within Excel. Accordingly, the user interface module (31) detects the input event and stores the input natural language request in an internal buffer. Additionally, the user interface module (31) collects selection information of objects such as cell ranges, shapes, images, and charts selected by the user in the Excel sheet simultaneously with the above input.

[0075] In the case of the selected object detection and reference information collection function, the user interface module (31) is linked with the cell reference selection detection module (32) to automatically detect the type of object (cell, shape, image, chart, etc.) selected by the user. At this time, the detected object information may include the object type, address or reference range, format and data attributes, chart series information (SERIES formula), etc. This information is subsequently converted into structured data in JSON format and transmitted to the AI ​​execution plan generation module (50).

[0076] In the case of the command structuring function, the user interface module (31) combines natural language request statements and object reference information to structure them into AI execution data (AIExecData). The structured data is in the form of JSON or internal command objects and may include the command purpose (action), input data range (range), execution condition (parameter), output format (output type), etc. The structured AIExecData is a data format prepared so that the AI ​​execution plan generation module (50) can subsequently establish an execution procedure.

[0077] In the case of the processing request transmission function, the user interface module (31) transmits structured command data to the AI ​​execution plan generation module (50) via an internal communication interface (API or event bus).

[0078] In the case of the result reception and feedback display function, when the result of the AI ​​execution plan generation module (50) or the AI ​​analysis module (42) is returned, the user interface module (31) visually displays it on the Excel screen. For example, the result display method may consist of one or more of automatic result cell entry, output of result message or chart summary, automatic generation of recommended functions / formulas, and visual highlighting of the result area.

[0079] The operation flow of the above-described user interface module (31) may be as follows. First, there is a user input step. In the user input step, the user inputs a command into the natural language input window within the Excel TaskPane and simultaneously selects a cell or object to be referenced. Accordingly, the user interface module (31) detects the input event and automatically recognizes the selected object information (cell, shape, chart, etc.). In the case of a chart, the related SERIES formula and reference range are detected together. Subsequently, there is an input detection and object recognition step. In the input detection and object recognition step, the user interface module (31) combines the detected object information and the user input statement and converts it into a JSON structure in the form of, for example, "{action, target_range, chart_info, user_intent}". Subsequently, there is a command transmission step. In the command transmission step, the user interface module (31) transmits the structured data to the AI ​​execution plan generation module (50). Accordingly, the establishment of an execution procedure by the AI ​​execution plan generation module (50) is triggered. Subsequently, there is an AI processing and result reception step. In the AI ​​processing and result reception stage, the output value of the AI ​​analysis result or execution plan is transmitted to the user interface module (31), and the result is displayed in a cell or TaskPane within the Excel screen.

[0080] According to the user interface module (31) described above, the user can perform data processing and analysis using natural language commands without writing complex formulas, macros, or VBA. Additionally, cell or chart information selected by the user is automatically included, which increases the accuracy of AI analysis and reduces unnecessary input steps. Furthermore, execution results are displayed in real-time within Excel, allowing the user to intuitively check the operation process of the AI. Moreover, since the user interface module (31) can be linked with other analysis tools such as Google Sheets and Power BI in addition to Excel, it has high scalability to future cloud-based analysis environments.

[0081] The cell reference selection detection module (32) can detect selection events for objects (e.g., cells, shapes, pictures, charts, etc.) selected or specified by the user in the Microsoft Excel environment in real time, and classify the selection information by object type to perform a preprocessing function to convert it into structured data that can be used for AI analysis or establishing an execution plan.

[0082] That is, the cell reference selection detection module (32) operates in conjunction with the basic workspace (UI Layer) of Excel and can extract the location, data range, formatting properties, reference cell address, etc. of the selected object and convert them into a form (JSON, internal data object, etc.) that can be referenced by the subsequent AI execution plan generation module (50) or AI analysis / response processing module (81).

[0083] This cell reference selection detection module (32) can be described as a preprocessing interface that converts the user's visual selection behavior into a data input signal that the AI ​​can understand.

[0084] The main functions of the cell reference selection detection module (32) may include an object selection event detection function, a selected object type classification function, a cell reference data extraction function, a chart reference area recognition function, an object data conversion (JSON encoding) function, a selection information combination function, and a state management and event log recording function.

[0085] For the object selection event detection function, when a user selects a cell, chart, shape, picture, etc. on an Excel sheet, the corresponding event is detected via the Excel API (SelectionChange Event, etc.) and immediately registered in the module's internal queue.

[0086] For the selected object type classification function, it determines whether the detected selected target is a general cell, a shape / image object, or a chart object, and passes it to the respective classification routine.

[0087] For the cell reference data extraction function, when a specific cell is selected, it reads the address, data value, and formatting (number format, color, border, etc.) of the selected cell, and includes merge range or named range information if necessary.

[0088] In the case of the chart reference area recognition function, when a specific chart object is selected, it analyzes the series formulas and data ranges referenced by the chart to extract the reference cell addresses for each series.

[0089] In the case of the object data conversion (JSON encoding) function, the extracted data is converted into JSON or an internally structured object (e.g., {"type":"cell", "address":"B2:B10", "value":[1,2,3], "format":"number"}) so that the AI ​​analysis module (42) can utilize it.

[0090] In the case of the selection information combination function, the detected object information is combined with the user natural language request received by the user interface module (31) and integrated into AI Analysis Request Data.

[0091] For the state management and event log recording functions, logs such as the time of the selected event occurrence, the target object, and whether extraction was successful are stored internally or through a server (20) to ensure the possibility of auditing and recovery.

[0092] The operation flow of the cell reference selection detection module (32) above may be as follows.

[0093] First, a selection event detection step can be performed. In the selection event detection step, when a user clicks or drags one of the Excel default workspaces (e.g., cells, charts, shapes, etc.), the cell reference selection detection module (32) detects this in real time through Excel's SelectionChange event. If there is no selection or the cell is inactive, it branches to the No Selection path.

[0094] Next, a step for determining the type of selected object can be performed. In the step for determining the type of selected object, the cell reference selection detection module (32) reads the attributes of the selected object (e.g., Type, ObjectClass, etc.) and classifies it into one of “cell”, “shape / image”, or “chart”. Depending on the classification result, one of the cell detection submodule (not shown), image detection submodule (not shown), or chart detection submodule (not shown) is activated.

[0095] Next, the data extraction and structuring step can be performed. In the data extraction and structuring step, when a cell is selected, the cell address, data value, and formatting properties are extracted and structured. When a shape / picture is selected, image data is encoded, for example, in Base64 format. When a chart is selected, the SERIES formula and referenced cell range are analyzed and converted into a JSON structure.

[0096] Next, a JSON conversion and combination step can be performed. In the JSON conversion and combination step, all detected selection information is integrated into a JSON object. This information is combined with the natural language request received by the user interface module (31) and transmitted to the AI ​​analysis module (42) or the AI ​​execution plan generation module (50).

[0097] Finally, the data transmission and log recording steps can be performed. In the case of the data transmission and log recording steps, the completed selection information JSON is transmitted to the AI ​​execution plan generation module (50) via an internal data bus or API. At the same time, the detection and transmission history is recorded on the server (20).

[0098] According to the cell reference selection detection module (32) described above, the actual data range and attributes of the object selected by the user are included in the AI ​​analysis request, thereby enabling the AI ​​to establish an accurate execution plan based on context. Furthermore, the selection information is automatically recognized and the target of analysis is determined even if the user does not explicitly input "analyze this cell range." Additionally, by supporting various Excel object types such as shapes and charts in addition to cells, a general-purpose AI automation environment can be implemented. Moreover, since the input dataset required for AI analysis is automatically configured based on the Excel area selected by the user, a prior data specification process is unnecessary. Furthermore, both the selection event and the data transformation process are recorded, ensuring the auditability of the AI ​​analysis results. Additionally, the data structure (JSON-based) transmitted through the cell reference selection detection module (32) is standardized, making it easy to reuse in the higher-level AI execution plan generation and response processing stages.

[0099] The cell reference selection detection module (32) detects selection events corresponding to the user's selected objects (cells, shapes, pictures, charts, etc.) within the Excel environment, extracts attribute data by type of the selected object, converts it into a structured form, and then transmits it to the AI ​​analysis module (42) or the AI ​​execution plan generation module (50). Accordingly, the actual data area specified by the user in Excel is automatically reflected in the process of establishing the execution plan of the artificial intelligence, thereby improving the accuracy and efficiency of the analysis.

[0100] The reference history conversion module (33) converts reference information for a cell or area, image, or object (shape, chart, etc.) selected by the user within the Excel environment into a standardized data structure so that it can be used in the AI ​​execution plan generation process.

[0101] The reference history conversion module (33) receives reference events and meta-information collected from the cell reference selection detection module (32) and converts them into a format (JSON, Key-Value structure, internal object, etc.) that the AI ​​execution plan generation module (50) can understand. Through this, visual selections or range selections made by the user within Excel are connected to semantic reference entities that the AI ​​can process.

[0102] The main functions of the reference history conversion module (33) may include a reference event receiving function, a reference information normalization function, a reference metadata expansion function, an AI reference data packaging function, and a reference history management and delivery function.

[0103] In the case of the reference event receiving function, reference information (e.g., selected range address, sheet name, reference object type, etc.) transmitted from the cell reference selection detection module (32) is received in real time. If multiple reference events occur simultaneously, they can be processed sequentially by managing priorities or event queues.

[0104] In the case of the reference information normalization function, the received reference information is converted from an internal Excel representation (e.g., "Sheet1!B2:C10") into a standard reference format recognizable by the AI ​​execution plan generation module (50) or the AI ​​analysis / response processing module (81). At this time, it is normalized to include cell addresses, column / row indices, reference types (single cell, range, table, chart, etc.), and data types (INT, STRING, DATE, etc.).

[0105] Since the reference information must be converted into a logical plan-making unit that the AI ​​can understand, the normalized reference information can be used as the main input to the AI ​​execution plan generation module (50).

[0106] Depending on the needs, the reference information may be used directly for analysis or utilized in the question-and-answer process. For example, when a user says, "Tell me the average of this column," the referenced range may be normalized and used directly for analysis. In this case, the AI ​​analysis / response processing module (81) may also be a recipient.

[0107] For the reference metadata extension function, data attributes of the referenced area (e.g., value, existence of formula, field name estimation, number of data, etc.) are collected as auxiliary data and added as metadata. The extension field is configured so that it can be transmitted to the AI ​​analysis module (42) or the AI ​​execution plan generation module (50) if necessary.

[0108] For the AI ​​reference data packaging function, normalized and extended reference information is packaged in the AIExecData or AIPlanData format. The normalized and extended reference information is assigned a unique ID and is designed to maintain reference consistency during subsequent AI response processing steps.

[0109] For the reference history management and delivery function, the converted reference history is delivered to the AI ​​execution plan generation module (50) via an internal data bus or method call. At the same time, the recent reference history is shared with the user interface module (31) to allow the user to check or reuse previous selections.

[0110] The operation flow of the above reference history conversion module (33) may be as follows.

[0111] First, a reference detection step can be performed. When a user selects a cell or area in an Excel sheet, the cell reference selection detection module (32) generates the corresponding event. The selection information (e.g., sheet name, cell address, object ID) is transmitted to the reference history conversion module (33).

[0112] Next, a step of receiving reference information can be performed. The reference history conversion module (33) receives reference data detected through an event listener and stores it in an internal buffer.

[0113] Next, a reference structure conversion step can be performed. The reference history conversion module (33) converts the Excel native representation into a standard structured format (JSON, etc.). The attributes of the reference target (number of cells, data type, value distribution, etc.) are merged to form a single reference object.

[0114] Next, the metadata expansion step can be performed. Additional information, such as field names, data patterns, and missing values ​​in the selected area, is automatically analyzed and included in the reference object.

[0115] Next, the AI ​​data packaging step can be performed. The final converted reference history is packaged into AIExecData or AIPlanData and delivered to the AI ​​execution plan generation module (50).

[0116] Finally, the transmission and feedback steps can be performed. When the transformation result is successfully transmitted, the user interface module (31) visually indicates that the corresponding reference area has been reflected in the AI ​​analysis. In the event of an error (e.g., invalid reference, selection of an empty cell, etc.), the reference history transformation module (33) returns a warning message to the user interface module (31). The above errors can be detected during the normalization of reference information and metadata expansion process.

[0117] According to the reference history conversion module (33) described above, by structuring selection information such as cells, tables, and charts, the AI ​​execution plan generation module (50) or the AI ​​response processing module (80) can understand the data within Excel in a structured manner. Additionally, since the reference information is standardized, it is easy to link with other components such as the AI ​​execution plan generation module (50) or the AI ​​response processing module (80). Furthermore, by converting the user's visual selection (such as a cell click) into a data semantic unit rather than a simple event, natural language commands and data context are combined. Moreover, the selection history and conversion format are managed in an integrated manner, significantly reducing the possibility of errors such as incorrect references or missing ranges. Additionally, conversion and visual feedback are provided immediately upon a change in reference, allowing the user to intuitively verify the reference range recognized by the AI.

[0118] In FIG. 2, the request analysis and classification module (40) receives user request data from the user interaction interface module (30), analyzes the intent, action, and context of the natural language command, converts the analysis results into structured command interpretation information, and determines the corresponding execution type based on the analysis results and outputs it as an AI execution judgment result. Here, the AI ​​execution judgment result may include the intent of the command, the action of the command, a reference, an execution type, and structured command interpretation information. Here, the intent of the command is information indicating what purpose the user has when issuing the command, such as, for example, "Calculate the sum" (Intent: Summarize). The action of the command indicates what action the user is requesting, such as, for example, "Add a trendline" (Action: Add Trendline). The reference is information identifying which data object (cell, range, table, chart, etc.) the user's command targets, such as, for example, "Sort based on the first column of this table" (Reference: Table 1, First Column). The execution type is the result of a decision determining how to execute a user's command, and may include, for example, formula-based (e.g., using Excel formulas), VBA script-based (e.g., macro automation), or data query-based (e.g., processing SQL-like queries). Structured command parsing information refers to the result of converting analyzed information (including the command intent, command behavior, references (context), and execution type) into a machine-executable syntactic structure or command tree form.

[0119] In other words, the request analysis and classification module (40) analyzes user input (text or voice-converted command) to classify it into conversational requests, information search requests, task execution requests, etc., and analyzes the intent, action, and context of the command.

[0120] The request analysis and classification module (40) utilizes a pre-trained language model (LLM) or a rule-based parser to interpret commands into a structure that can be converted into Excel formulas, VBA, SQL, etc., and determines an execution path (formula-based, VBA-based, etc.) suitable for processing the request.

[0121] That is, the request analysis and classification module (40) analyzes natural language commands or voice text input by the user to identify the intent and context of the command, and classifies them into information requests, conversational requests, task execution requests, etc. The operation path of the subsequent processing module is determined according to the classified request type, and the request analysis and classification module (40) improves the analysis accuracy by breaking down and reconstructing complex requests into sub-command units.

[0122] In addition, the request analysis and classification module (40) utilizes a language model-based analysis, rule-based parsing, or hybrid analysis engine to precisely interpret the user's command structure and context, thereby realizing an advanced understanding of the request.

[0123] The operation flow of the request analysis and classification module (40) may be as follows. Natural language text of a user command is received, and the request type, execution purpose, target scope, etc. of the command are analyzed through a pre-trained LLM or parser. Then, based on the analysis results, the request type (conversation, information search, task request), intent analysis (action to be executed, calculation purpose, target data, etc.), and execution method (formula execution, VBA macro, mixed method) are determined (judged). The determined execution method and analysis information are transmitted to the AI ​​execution plan generation module (50) and the task request execution processing branch module (61) of the job execution flow control module (60).

[0124] The request analysis and classification module (40) may include a request history classification module (41), an AI analysis module (42), and an AI judgment module (43).

[0125] The request history classification module (41) classifies natural language commands or processing requests received from a user by intent and request category, and can transmit the classification results to the AI ​​execution plan generation module (50) or the AI ​​response processing module (80).

[0126] That is, the request history classification module (41) is a functional gateway for determining what needs to be done, and can identify whether a natural language command is a simple query (e.g., "What is the average of this column?"), a data processing request (e.g., "Filter only values ​​greater than 0 in column B"), a visualization request (e.g., "Show this data as a chart"), and can automatically determine the path for establishing a subsequent AI execution plan based on this.

[0127] The main functions of the request history classification module (41) may include request semantic interpretation and intent classification functions, complex request decomposition functions, reference linkage verification functions, execution path determination and routing functions, and request log and feedback recording functions.

[0128] For the request semantic interpretation and intent classification function, natural language input (and reference-transformed data) transmitted from the user interface module (31) is received, and the semantic type of the command is determined through a built-in intent classifier or rule-based parser. For example, if the command is "calculate the average" or "calculate the sum," it can be determined as a calculation type request. If the command is "show only values ​​greater than or equal to 0" or "keep only rows where column A is 'sales'," it can be determined as a filtering / conditional type request. If the command is "make a bar graph," it can be determined as a visualization type request. If the command is "what is the largest value in this data?", it can be determined as a question-answer type request.

[0129] In the case of the complex request decomposition function, if multiple requests are included within a single command (e.g., "Calculate the average of this column and make it into a table"), the command is decomposed into multi-stage execution units (sub-tasks) to independently identify each request type. The decomposed requests are individually transmitted to the AI ​​execution plan generation module (50), which includes execution order information.

[0130] In the case of the reference linkage verification function, the data reference object transmitted from the reference history conversion module (33) is compared with the request content to verify whether the data range specified by the user matches the request content. For example, when a request for "average of column B" is made, it checks whether the reference object is column B and returns a warning message to the user interface module (31) if there is a mismatch.

[0131] In the case of the execution path determination and routing function, depending on the classified request type, it is determined whether the request is to be passed to the AI ​​execution plan generation module (50) or a simple response to be directly performed by the AI ​​response processing module (80) (e.g., FAQ, explanatory answer, etc.) and routed.

[0132] For the request log and feedback recording functions, metadata such as classification results, processing targets, and execution paths is stored in internal logs, and feedback data is accumulated so that it can be utilized for AI model performance improvement or usage pattern analysis.

[0133] The operation flow of the above request history classification module (41) may be as follows.

[0134] First, the request input reception step can be performed. Natural language commands and reference data (including cell selection information) are received from the user interface module (31).

[0135] Next, the intent analysis and type determination steps can be performed. The structure of the command sentence is interpreted through a built-in NLP engine or a rule-based analyzer, and the intent and category (which can be referred to as type or target) of the command are determined.

[0136] Next, the complex request decomposition step can be performed. If multiple task elements exist within a single command, the command is separated by execution stage, such as "calculation → visualization."

[0137] Next, the reference information linkage and verification step can be performed. By utilizing the reference object provided by the reference history conversion module (33), it is checked whether the requested data range is valid.

[0138] Next, the execution module routing step can be performed. If the request is of the data processing or analysis type, it is forwarded to the AI ​​execution plan generation module (50), and if it is of the simple query type, it is forwarded directly to the AI ​​response processing module (80).

[0139] Finally, the log recording and feedback provision steps can be performed. The classification results and processing paths are recorded in an internal log, and the results are fed back to the user interface module (31) to indicate the status of the request, such as whether it is recognized or cannot be processed.

[0140] According to the request history classification module (41) described above, complex user commands are automatically subdivided so that the AI ​​can process them step by step, thereby enabling stable automation of complex tasks. Additionally, branching to the AI ​​execution plan generation module (50) and the AI ​​response processing module (80) is automatically performed depending on the nature of the request, thereby improving system efficiency and processing consistency. Furthermore, data processing errors can be prevented in advance by verifying in real-time whether the referenced cell range and the command content match. Additionally, the type of request recognized and which module is processing can be visually displayed, allowing the user to intuitively understand the system's operational status. Moreover, classification and routing logs are accumulated and can be utilized for improving AI intent classification accuracy and recommending user-customized commands in the future.

[0141] The AI ​​analysis module (42) is a core preprocessing engine that receives natural language commands or request sentences input through the user interface module (31) and analyzes the intent and context in a multi-layered manner.

[0142] The AI ​​analysis module (42) can convert natural language commands into executable structured representations (e.g., SQL, Excel formulas, VBA code, data processing logic, etc.) by utilizing a combination of a pre-trained language model (LLM) and a rule-based parser. The AI ​​analysis module (42) structures the converted analysis results into an initial form of an AI execution schema and then transmits them to the AI ​​judgment module (43) and the AI ​​execution plan generation module (50). Here, structuring into an initial form of an AI execution schema means structuring the purpose, target, and context of the command to provide basic data for planning.

[0143] The main functions of the AI ​​analysis module (42) may include a command intent analysis function, a reference context identification function, a function to convert language into an execution expression, an execution schema configuration function, and a delivery and verification interface function.

[0144] In the case of the command intent analysis function, the user's natural language command is interpreted based on LLM to extract the purpose of the request (e.g., data retrieval, calculation, summarization, transformation, etc.).

[0145] In the case of the reference context identification function, actual objects such as execution target cells, tables, and charts can be recognized by combining the reference target information transmitted from the cell reference selection detection module (32) and the reference history conversion module (33).

[0146] For the function of converting language into execution expressions, natural language commands can be converted into SQL, Excel formulas, VBA code, etc., through a hybrid conversion method of rule-based parsers and LLMs.

[0147] For the execution schema configuration feature, an AI execution schema can be generated based on analysis results. This AI execution schema may include metadata such as the request purpose and type, reference data and scope, execution conditions and constraints, target locations for result reflection, and metadata for logs and restoration.

[0148] In the case of the transmission and verification interface function, the generated AI execution schema can be transmitted to the AI ​​judgment module (43) and the AI ​​execution plan generation module (50), and a request for verification of execution feasibility can be performed in parallel.

[0149] The operation flow of the AI ​​analysis module (42) can be as follows.

[0150] First, a command reception step can be performed. A natural language command is entered through a user interface module (31) (e.g., TaskPane UI, dialog box, etc.). After being classified as a command request through a request history classification module (41), it is transmitted to an AI analysis module (42).

[0151] Next, an intent and reference analysis step can be performed. After extracting the intent, action, and context of the command from the input sentence, reference information such as cells, tables, and charts provided by the cell reference selection detection module (32) and the reference history conversion module (33) is combined. Here, combining reference information refers to the process of combining the intent and action extracted from the user's natural language command with the contextual reference (i.e., actual Excel object information) detected by other modules into a single Integrated Semantic Map. The reference information combined by the AI ​​analysis module (42) can be transmitted from the cell reference selection detection module (32) and the reference history conversion module (33). The AI ​​analysis module (42) can receive the ID, address, range coordinates, etc., of an object such as a cell, range, shape, or chart selected by the user with a mouse from the cell reference selection detection module (32). Additionally, the AI ​​analysis module (42) may receive information from the reference history conversion module (33) that converts the detected reference object into a standardized data structure (e.g., JSON, XML). The reference metadata transmitted from these two modules is combined within the command parse tree structure inside the AI ​​analysis module (42). That is, the reference metadata is combined with the target and context fields of the natural language parse result structure (command parse tree), and this combined result is mapped to the input reference field (inputRef) of the AI ​​execution schema and subsequently transmitted to the AI ​​judgment module (43) and the AI ​​execution plan generation module (50).In other words, reference information is not merely data referenced by the analysis module, but rather acts as an element that completes the inputRef and context.binding fields within the AI ​​execution schema by being combined with the action and context structures of commands extracted from natural language sentences.

[0152] Next, command conversion and schema formation steps can be performed. The AI ​​analysis module (42) can convert natural language commands into an executable form (SQL, function expression, VBA, etc.) using LLM-based language analysis and a rule-based parser. The conversion result is organized into an AI execution schema structure and includes detailed parameters required for execution control.

[0153] Finally, output and linkage steps can be performed. The conversion results are transmitted to the AI ​​judgment module (43) to determine execution suitability and conditions. Subsequently, the AI ​​execution schema is transmitted to the AI ​​execution plan generation module (50) and materialized into an actual execution procedure. Logs of the analysis process are recorded in the system integration management module (90) to ensure traceability.

[0154] According to the AI ​​analysis module (42) described above, even ambiguous requests based on natural language can be converted into clear execution information (functions, SQL, VBA, etc.) to accurately support the automated execution of user commands. Additionally, the degree of contextual command processing can be improved by simultaneously analyzing various reference targets, such as cells, shapes, and charts selected by the user. Furthermore, by structuring and storing all analysis results in the form of an AI execution schema, rapid restoration and verification are possible in the event of an execution error. Moreover, by providing analysis results in a consistent data structure, the interface efficiency between the AI ​​judgment module (43) and the AI ​​execution plan generation module (50) can be maximized. Additionally, since automation can be realized that allows for immediate execution using only natural language commands without the need to directly input complex formulas or macro commands, the user experience can be improved.

[0155] The AI ​​judgment module (43) receives the natural language interpretation results (command intent, execution target, reference data, execution schema, etc.) transmitted from the AI ​​analysis module (42) as input, and determines and decides a suitable execution path (Execution Path) to perform the corresponding task request.

[0156] That is, the AI ​​judgment module (43) can determine one or more execution options among a formula-based execution path, a VBA-based execution path, and a hybrid path by comprehensively considering the characteristics of the request (e.g., data processing type, formula calculation type, macro automation type, visualization type, etc.), data structure, Excel environment settings, user preferences, and past execution patterns based on the analysis results of the AI ​​analysis module (42).

[0157] The execution path determined in this way is transmitted to the task request execution processing branch module (61), where it is guided to the user as an execution option or automatically selected and transmitted to the execution flow control and restoration module (72). The execution flow control and restoration module (72) subsequently ensures a stable and restoreable execution process by integrally controlling procedures such as pre-execution verification, execution control, exception handling, and result reflection.

[0158] The main functions of the above AI judgment module (43) may include an execution path determination function, an execution condition evaluation function, a multi-path option suggestion function, a user preference-based selection function, and an execution path delivery function.

[0159] The execution path determination function can select the most suitable processing engine (formula / VBA) according to the type of command (calculation, iteration, conditional branching, input / output, visualization, etc.) by reviewing the execution schema and command parameters generated by the AI ​​analysis module (42).

[0160] The execution condition evaluation function can verify execution feasibility by evaluating conditions such as Excel environment settings, external API availability, VBA macro security policies, and data size.

[0161] The multi-path option suggestion feature can generate multiple execution path candidates when multiple execution methods are possible (e.g., the same calculation can be performed using a formula or VBA), evaluate the characteristics of each option (speed, scope of automation, maintainability, etc.), and present them to the user.

[0162] The user preference-based selection function can refer to past user execution logs from the system integration management module (90) and apply the execution type that the user previously preferred first.

[0163] The execution path delivery function can deliver the determined execution path and related metadata (e.g., execution method, required resources, estimated processing time, etc.) to the task request execution processing branch module (61) or the execution control module (64).

[0164] The operation flow of the AI ​​judgment module (43) described above may be as follows.

[0165] First, an input receiving step can be performed. The AI ​​judgment module (43) receives natural language interpretation results and execution schemas (121) from the AI ​​analysis module (42). The input data may include the intent of the command, the context of execution, cell reference information, function requirements, data range information, etc.

[0166] Next, a task type analysis step can be performed. The AI ​​judgment module (43) analyzes the command structure to determine whether the task can be processed based on formulas, requires repetitive macro automation, or requires a combination of both methods.

[0167] Next, an execution path evaluation step can be performed. The AI ​​judgment module (43) comprehensively reviews the Excel environment status, whether VBA is available, the size of the processing target (e.g., whether there are thousands of rows of data), user permissions, etc.

[0168] Next, the execution path determination step can be performed. The AI ​​judgment module (43) determines the most suitable execution method (formula / VBA / mixed type). If there are multiple possible executions, the AI ​​judgment module (43) evaluates the expected efficiency and reliability of each path and assigns a priority.

[0169] Finally, the execution path transmission step can be performed. The AI ​​judgment module (43) transmits the finally determined execution path to the task request execution processing branch module (61) to propose it to the user, or transmits it directly to the execution control module (64) to initiate execution.

[0170] According to the AI ​​judgment module (43) described above, the most efficient execution method is automatically selected based on the structure and size of the data and the system environment for the same natural language request, thereby preventing unnecessary resource waste and maximizing execution efficiency. Additionally, the level of business automation is improved because the AI ​​suggests the most appropriate execution method or automatically executes it without the user having to decide whether to use a formula or VBA. Furthermore, since execution conditions are evaluated in advance, the possibility of failure due to VBA security errors or function incompatibility can be prevented in advance. Additionally, customized execution options are presented based on user history and preferences, thereby increasing the efficiency of repetitive tasks and user satisfaction. Moreover, through the three-stage linkage structure of the AI ​​analysis module (42), the AI ​​judgment module (43), and the task request execution processing branch module (61), natural language commands are smoothly converted into actual executable code or formulas, and the logical consistency of the entire system can be maintained.

[0171] In FIG. 2, the AI ​​execution plan generation module (50) can generate an AI execution schema by receiving an AI execution judgment result from the request analysis and classification module (40) and specifying natural language commands into actual execution procedure units.

[0172] In other words, the AI ​​execution plan generation module (50) receives command semantic information (intention, action, target, etc.) interpreted by the AI ​​analysis module (42) and an execution path (e.g., formula-based, VBA-based, SQL-based, etc.) determined by the AI ​​judgment module (43) as input, and can configure this into a specific execution scenario (i.e., a complete AI execution schema) that can be executed in actual Excel or an external linked system.

[0173] That is, the AI ​​execution plan generation module (50) can automatically generate an execution scenario that defines what, in what order, and in what way to execute by combining the analysis results from the AI ​​analysis module (42) (e.g., user intent, type of behavior, target object, condition, etc.) and the execution judgment results from the AI ​​judgment module (43) (e.g., execution direction or strategy (e.g., formula application, macro call, table transformation, SQL execution, etc.)).

[0174] The AI ​​execution plan generation module (50) goes beyond simple command interpretation to configure (generate) an AI execution schema including execution flow, conditions, dependencies, error recovery paths, etc., and outputs it in a structured form so that the execution control module (64) and the result reflection module (65) can execute it as is.

[0175] The main functions of the AI ​​execution plan generation module (50) may include command structure analysis and procedure separation functions, execution step design functions, condition and constraint processing functions, AI model linkage and pipeline definition functions, and AI execution schema generation and structuring functions.

[0176] In the case of the command structure interpretation and procedure separation function, the AI ​​execution plan generation module (50) interprets the structure of the command based on the user's natural language command into logical units of "Action + Target + Condition". For example, the command "Summarize this table and show it as a chart" can be separated into two execution steps [Table Summary → Chart Creation], and each step is designed as an independent and sequential execution unit.

[0177] For the execution step design function, the AI ​​execution plan generation module (50) automatically configures execution steps that can be performed in Excel based on the interpreted command structure. Each step is defined in units of Excel API calls or AI model calls and can follow the order of data preparation → analysis → visualization → result generation. This enables consistency and automation of processing for multi-step commands.

[0178] For the condition and constraint processing function, the AI ​​execution plan generation module (50) adjusts the processing logic of each execution step by reflecting data constraints (e.g., filter conditions, data type restrictions, range selection, etc.) transmitted from the reference history conversion module (33) and the request history classification module (41). For example, a condition such as "analyze only rows where sales are 1,000 or more" is automatically converted into a filtering execution step.

[0179] For the AI ​​model integration and pipeline definition functions, certain execution steps—which refer to tasks such as structurally interpreting user natural language commands, performing data analysis, determining visualization methods, or semantically inferring unspecified conditions—define integration with AI engines like LLM, data analysis models, and chart recommendation models, and include their input / output flows in the execution plan.

[0180] In the case of the AI ​​execution schema generation and structuring function, the execution plan completed through the preceding process is structured into an AI execution schema. This AI execution schema is generated as a structured structure (JSON / XML, etc.) including the purpose, input, output, sequence, error handling path, model linkage information, etc. of each step, and is transmitted to the task execution flow control module (60) and the AI ​​response processing module (80) to be used for subsequent execution.

[0181] The operation flow of the above AI execution plan generation module (50) may be as follows.

[0182] First, the step of receiving input information can be performed. The AI ​​execution plan generation module (50) can receive information on the intent and behavior of a user command from the request history classification module (41), and information on the reference target (cell, table, chart, etc.) to which the command will be applied from the task execution target selection module (62).

[0183] Next, command analysis and step decomposition steps can be performed. The AI ​​execution plan generation module (50) interprets natural language commands and decomposes them into detailed execution steps (e.g., sorting, filtering, calculation, visualization, etc.), and if necessary, designs the execution flow by inferring implicit semantics through LLM. For example, if the above natural language command is interpreted as an execution command (e.g., "sort this table," "calculate the average of column A," "duplicate this sheet"), the task execution flow control module (60) will output task execution result data (e.g., formulas, cell values, charts, tables, etc.) through the following steps.

[0184] Next, the steps of setting execution conditions and reflecting constraints can be performed. The AI ​​execution plan generation module (50) identifies explicit conditions included in the command (e.g., "top 10", "1,000 or more") and data constraints transmitted from external modules and reflects them as conditions for each execution step. Here, the external modules mainly include a reference history conversion module (33), a request history classification module (41), and, in some cases, a task execution target selection module (62). Based on the reference information and condition information transmitted from these modules, the AI ​​execution plan generation module (50) sets execution conditions and constrains the execution flow. For example, the reference history conversion module (33) can structure and transmit reference objects such as cells, ranges, tables, and charts selected by the user within Excel, the request history classification module (41) can extract and transmit the intent, conditions, and operation requirements of the user command, and the task execution target selection module (62) can transmit information on the range or type (e.g., tables, columns, sheets) of the actual execution target object.

[0185] Next, the execution scenario configuration step can be performed. Each execution step is logically connected and sequenced, and an execution flow control structure, such as parallel execution or conditional branching, is defined. If model calls are included, API parameters, response structures, etc., are also included in the execution scenario. Here, each execution step is an individual work unit derived by the AI ​​execution plan generation module (50) to execute user commands, and may consist of a data processing step (data filtering, sorting, grouping, conditional search, removal of missing values, etc.), a formula / calculation step (basic operations such as formula insertion, sum, average, etc., and performing complex calculation logic), a data analysis step (statistical analysis (variance, correlation coefficient, etc.), trend prediction, extraction of top items, aggregation by category), a visualization step (chart recommendation, chart creation, chart styling), a response configuration step (generating summary results, outputting explanatory sentences, configuring data for user responses), and an AI model call step (LLM call (natural language interpretation, response generation, etc.), chart recommendation engine call, analysis model call). These steps are connected in a logical order or in parallel / conditional flow to configure the execution scenario.

[0186] Finally, the steps for generating and outputting the AI ​​execution schema can be performed. The configured execution scenario is packaged into an AI execution schema, and the AI ​​execution schema (i.e., the structured execution scenario data structure generated by the AI ​​execution plan generation module (50)) is generated in a format that can be processed as is by the execution control module (64), the result reflection module (65), etc. This AI execution schema clearly includes the purpose, input, processing method, output, error handling flow, etc., of each step.

[0187] According to the AI ​​execution plan generation module (50) described above, the reliability and repeatability of AI-based automation are enhanced by converting the user's abstract or complex natural language commands into structured Excel execution logic. Additionally, by subdividing and structuring the execution procedure, fault tolerance is secured so that even if an error occurs at a specific stage during execution, the entire process is not interrupted and only a part of it can be re-executed. Furthermore, since a pre-execution plan is established based on the command analysis results, the subsequent AI response processing module (80) can be executed immediately without unnecessary context reinterpretation, and the overall processing speed is improved. Moreover, since the structure of the execution plan is standardized, various AI models (e.g., language models, analysis engines, visualization engines, etc.) can be flexibly replaced or expanded, resulting in excellent maintainability and scalability of the system. Additionally, the execution plan can be visually displayed to the user, providing transparency and predictability regarding command processing by clearly showing how each step will be performed.

[0188] In FIG. 2, the task execution flow control module (60) controls actual Excel operations based on an AI execution plan (i.e., an AI execution schema) from the AI ​​execution plan generation module (50), verifies security status or cell protection before execution, sets the execution target location (new sheet or existing cell), executes Excel formulas, VBA code, etc., and reflects the results. The AI ​​execution schema is a structured execution procedure generated by the AI ​​execution plan generation module (50) based on the user's natural language commands. The AI ​​execution schema is a data structure designed to be interpretable and executable by the task execution flow control module (60) responsible for actual execution. Therefore, controlling execution based on an execution plan from the AI ​​execution plan generation module (50) means receiving the AI ​​execution schema and performing execution control accordingly.

[0189] The operation flow of the task execution flow control module (60) may be as follows. The task execution flow control module (60) receives an AI execution schema from the AI ​​execution plan generation module (50) and checks the status of the sheet and cell to which the received AI execution plan will be applied. For example, it may check whether the sheet is protected, the cell lock status, whether a pivot table exists, and whether VBA macros are allowed. If the check passes, the task execution flow control module (60) performs branch processing according to the execution plan (e.g., execution of Excel formulas, execution of VBA automation, etc.) and outputs the result to a designated location (a new sheet or a specific cell). The output result may include calculated values, inserted formulas, generated charts, modified tables, etc.

[0190] The task execution flow control module (60) may include a task request execution processing branch module (61), a task execution target selection module (62), a pre-execution verification module (63), an execution control module (64), and a result reflection module (65).

[0191] The task request execution processing branch module (61) can determine which processing routine to use to execute the request based on the execution plan information transmitted from the AI ​​execution plan generation module (50) or the request data transmitted from the request history classification module (41), and can automatically branch to a corresponding execution submodule (e.g., Excel formula builder, VBA executer, data processor, etc.).

[0192] That is, the task request execution processing branch module (61) acts as a control gateway for the execution stage that determines how the user's natural language request will be executed in actual Excel.

[0193] The main functions of the task request execution processing branch module (61) may include an execution path determination function, a processing routine mapping function, an execution request transmission and monitoring function, an error detection and alternative path processing function, a result integration and response delivery function, etc.

[0194] The execution path determination function analyzes the execution type code (action_type) included in the execution plan data (e.g., JSON or internal execution schema) transmitted from the AI ​​execution plan generation module (50) to determine which type the corresponding task is among Formula Generation, VBA Macro Execution, Data Processing, Charting / Visualization, and Response (Text-based AI response provision).

[0195] The processing routine mapping function selects an internal processing routine corresponding to the determined execution type and maps a corresponding routine linked to that internal processing routine. Here, the internal processing routine is a basic processing branching logic defined within the task request execution processing branching module (61), which determines the direction of execution for each request type (formula, data processing, AI analysis, etc.). That is, the internal processing routine analyzes the input request and determines which type of processing module to forward it to, and can be described as a higher-level branching routine that selects and maps a corresponding routine. The corresponding routine can be described as an execution function or process within an actual processing module (e.g., AI analysis module (42), AI execution plan generation module (50), etc.) connected as a result of the internal processing routine's decision, and processes the request through actual execution, interpretation, planning, etc. The corresponding routine can be described as a lower-level execution unit that is called and forwarded by the internal processing routine.

[0196] The execution request transmission and monitoring function transmits an execution request object (i.e., a branched processing request) generated by an internal processing routine to a corresponding routine, monitors the execution status (waiting, running, completed, error, etc.), and transmits the status value to a user interface module (31) to provide real-time feedback to the user. Here, the branched processing request may refer to an execution request data object (Execution Request Object) selected and determined by the processing routine mapping function. For example, the branched processing request refers to execution unit request data generated by combining the result of request history classification, the result of internal processing routine determination, and reference target information.

[0197] The error detection and alternative path processing function, when an error is detected during the execution of each processing routine, executes a recovery scenario for the exception situation or transmits an error message to the AI ​​response processing module (80) to provide guidance to the user. For example, the error message may be "The selected cell range is invalid," "Reference data was corrupted during chart creation," etc.

[0198] The result integration and response delivery function integrates the results generated in each processing routine (e.g., calculated values, generated charts, text responses, etc.) into a common result format and delivers them to the AI ​​response processing module (80) or user interface module (31).

[0199] The operation flow of the above-mentioned task request execution processing branch module (61) may be as follows.

[0200] First, the step of receiving execution requests can be performed. Execution target requests and execution plan data are received from the AI ​​execution plan generation module (50) or the request history classification module (41).

[0201] Next, an execution type determination step can be performed. By analyzing the execution type field of the received data, it is determined whether the request is of the formula generation type, data processing type, macro execution type, or visualization type.

[0202] Next, the processing routine mapping and execution steps can be performed. An internal processing routine corresponding to the determined execution type is invoked, and the execution results are monitored in real time.

[0203] Next, the result collection and integration step can be performed. Result data is received from each processing routine and converted into a standardized result format.

[0204] Next, the result delivery and feedback step can be performed. The integrated result is delivered to the AI ​​response processing module (80) or user interface module (31) and displayed to the user in a visual or text-based manner.

[0205] Finally, logging and exception handling steps can be performed. All events of the execution process (requests, paths, results, errors, etc.) are logged, and the execution history is transmitted to an external server (20) if necessary.

[0206] According to the task request execution processing branch module (61) described above, formula creation, macro execution, data processing, etc., are automatically performed based on the execution path determined by the AI, so no separate manual operation is required after the user's command. In addition, since internal routines are dynamically selected according to each task type, new functions (e.g., data visualization, automatic report generation, etc.) can be easily added and expanded. Furthermore, through error detection and alternative path functions, stable processing is possible even with abnormal input or reference errors in the Excel environment. Also, by returning different results for each processing routine in a unified structure, subsequent response processing and UI display are maintained consistently. In addition, through execution status monitoring and result visualization functions, the user can intuitively check the execution process of the AI, and the reliability of the processing is increased.

[0207] The task execution target selection module (62) performs the function of identifying and determining specific objects (cells, tables, shapes, charts, etc.) within actual Excel that are the targets of the task execution, based on task type and reference range information derived from the AI ​​analysis results. Here, the AI ​​analysis results refer to the results of the AI ​​analysis module (42) and may include information such as, for example, the command intent, action type, context candidates, analyzed data range or cell location, and metadata required for execution (e.g., data type, function category, etc.).

[0208] That is, the task execution target selection module (62) ultimately determines what target to execute the command on, and is responsible for filtering and mapping referenceable data objects by reflecting the user's command intent (e.g., "Summarize this table", "Modify this chart").

[0209] The main functions of the task execution target selection module (62) may include an object mapping function, a task feasibility determination function, a multiple target management function, and a target object identification information output function.

[0210] The object mapping function searches for and maps corresponding actual objects within the internal structure of Excel based on reference information (cell address, sheet name, object ID, etc.) transmitted from the reference history conversion module (33). It supports various object types in Excel, such as tables, cell ranges, charts, shapes, and pictures.

[0211] The Task Availability Check function verifies whether the object is capable of performing the requested command type (e.g., "Summary," "Edit," "Delete," "Analysis"). For example, if the "Delete Chart" command is specified as a target for a table or cell, it is detected as an error. Here, the requested command may be structured user command data generated by the AI ​​analysis module (42), and after the execution method is determined through the AI ​​judgment module (43), it is passed to the task execution target selection module (62) to verify whether the target object can actually perform the command.

[0212] The Multi-target Handling feature assigns target objects sequentially or in parallel based on priority criteria (selection order, data type, explicit user mention, etc.) when multiple reference objects are involved.

[0213] The Target Context Packaging function structures the metadata of the selected object (e.g., name, type, sheet location, data range, etc.) and transmits it to the AI ​​execution plan generation module (50) or the AI ​​response processing module (80), etc.

[0214] The operation flow of the above task execution target selection module (62) may be as follows.

[0215] First, a step of receiving reference information can be performed. For example, the task execution target selection module (62) can receive the reference history processed by the reference history conversion module (33) as input (e.g., {Sheet1!A1:C10, "Sales Table"}).

[0216] Next, the Excel object mapping step can be performed. The task target selection module (62) searches for and matches actual objects through Excel's internal API or Office JS interface based on the received reference information.

[0217] Next, a target suitability verification step can be performed. The task execution target selection module (62) determines whether the type of object mapped to the task type (e.g., "analysis", "chart generation") transmitted from the request history classification module (41) is compatible.

[0218] Next, a step to determine the target for execution can be performed. The task execution target selection module (62) determines the primary target among multiple objects according to priority rules and, if necessary, adds secondary targets to the parallel processing list.

[0219] Finally, the target information packaging and delivery step can be performed. The task execution target selection module (62) transmits the structured information (JSON, XML, etc.) of the selected target object to the task request execution processing branch module (61) or the AI ​​execution plan generation module (50).

[0220] According to the task execution target selection module (62) described above, the discrepancy between the intent of the user command and the actual Excel object is reduced, thereby preventing errors where the command is applied to the wrong object. In addition, by providing a consistent mapping structure for various object types such as tables, charts, and shapes, the same command processing logic can be reused. Furthermore, since the AI ​​execution plan generation module (50) receives accurate target information, the model can set an accurate work path without performing unnecessary inference processes. Moreover, since the object selected by the user in the TaskPane UI matches the target object processed by the actual AI, the consistency of the user experience (UX) is improved.

[0221] The pre-execution verification module (63) can check in advance whether the execution plan transmitted from the AI ​​execution plan generation module (50) can actually be executed in an Excel environment, whether data references or operation conditions are valid, and whether there are no security or logical errors.

[0222] That is, the pre-execution verification module (63) can prevent execution errors, data corruption, abnormal processing, etc. in advance by automatically verifying cell references, function calls, data access rights, formula logic, data type consistency, etc. included in each step of the execution plan.

[0223] Accordingly, the pre-execution verification module (63) can operate as an intermediate inspection step between the AI ​​execution plan generation module (50) and the AI ​​response processing module (80), and returns execution permission (OK) or a request for correction (NG, Warning) based on the verification result, thereby improving the stability and reliability of the entire system.

[0224] The operation flow of the pre-execution verification module (63) may be as follows.

[0225] First, the execution plan reception step can be performed. The pre-execution verification module (63) receives a completed execution plan object (JSON, etc.) from the AI ​​execution plan generation module (50).

[0226] Next, a data reference check step can be performed. The pre-execution verification module (63) checks whether reference cells, ranges, tables, and name definitions exist within the actual Excel worksheet based on the information from the reference history conversion module (33).

[0227] Next, a step for checking formula and operation structures can be performed. The pre-execution verification module (63) verifies the syntactic and logical consistency of the operation expressions, formulas, and function calls within the plan. It checks the number of parameters, type compatibility, etc., when calling Excel APIs.

[0228] Next, security and authorization verification steps can be performed. The pre-execution verification module (63) checks whether the execution plan includes external access or sensitive file manipulation, and requests approval from the user (i.e., requests approval via the user interface module (31)) or blocks it if necessary.

[0229] Finally, a result summary and reporting step can be performed. The pre-execution verification module (63) classifies the verification results into "executable normally," "Warning," "Error," etc., and transmits them to the AI ​​response processing module (80), or in the event of an error, returns them to the AI ​​execution plan generation module (50) to request a replan.

[0230] According to the pre-execution verification module (63) described above, errors are detected in advance at the pre-execution stage, thereby minimizing the risk of Excel data corruption or program interruption. Additionally, the suitability of complex commands generated by the AI ​​is mechanically verified, thereby realizing highly reliable automatic processing without human user intervention. Furthermore, external integration of Excel, file access, and macro calls are controlled to block the possibility of malicious code execution or data leakage. Moreover, since "safe execution" is guaranteed through automatic diagnosis before execution rather than a method of correcting errors after they occur, the efficiency and satisfaction of the entire workflow are enhanced. Additionally, the pre-execution verification module (63) acts as a verification gate between the AI ​​execution plan generation module (50) and the AI ​​response processing module (80), thereby improving the quality of service (QoS) and maintainability of the AI-based automation system.

[0231] The execution control module (64) is a central management module that controls the flow of command execution within the actual Excel environment and executes, monitors, and terminates individual work units based on execution approval information transmitted from the AI ​​execution plan generation module (50) and the pre-execution verification module (63).

[0232] The execution control module (64) controls Excel's internal API, macro engine (VBA), or external add-in interface (Office JavaScript API, etc.) to execute the execution plan established by the AI ​​step by step, and can manage the progress status, results, and whether errors occur in real time for each step.

[0233] Accordingly, the execution control module (64) controls and coordinates the execution flow of the entire device, ensuring that the logical execution plan of the AI ​​is executed normally within the actual Excel environment.

[0234] The main functions of the execution control module (64) may include execution control management functions, execution scheduling and queue management functions, execution status monitoring functions, error handling and recovery functions, execution result integrated delivery functions, and logging and tracking management functions.

[0235] The Execution Flow Control function receives commands that have passed pre-execution verification and controls the execution order, conditions, and repetition of each command. It can manage single commands, complex commands (e.g., chaining of multiple functions), or parallel execution.

[0236] The Task Scheduling & Queueing function configures a task queue to process multiple commands sequentially and in parallel based on priority, even when multiple commands are requested simultaneously. Long-running tasks (e.g., large-scale data analysis) are processed asynchronously, and their status is monitored.

[0237] The Execution Status Monitoring function tracks the execution stages of each command (Start—In Progress—Complete—Error) and collects log and status information in real time. It records progress, processing time, and the location of errors.

[0238] The Error Handling & Recovery function stops and rolls back the operation or performs automatic retry logic if an error occurs during execution. It classifies the error type (missing data, formula error, permission denied, etc.) and provides feedback to the AI ​​response processing module (80) or the user.

[0239] The Result Integration & Reporting function aggregates result data from all execution units and transmits it to the AI ​​response processing module (80), and allows it to be visually reflected in an Excel sheet through the user interface module (31).

[0240] The logging and traceability function transmits execution logs, time information, execution targets, usage commands, etc., to a system integration management module (90) or an external server (20) to enable subsequent analysis and auditing.

[0241] The operation flow of the above execution control module (64) may be as follows.

[0242] First, the execution approval reception step can be performed. The execution control module (64) receives a "Verification complete (OK)" signal and an execution plan object from the pre-execution verification module (63). If there is an error or warning, the execution request is suspended or the correction step is called.

[0243] Next, the execution unit decomposition and registration steps can be performed. The execution control module (64) decomposes the commands within the execution plan into detailed execution units (Task Units) and registers them in an internal execution queue. For example, the execution control module (64) can configure a Task sequence in the order of {calculation → chart generation → result insertion}.

[0244] Next, execution scheduling and control steps can be performed. The execution control module (64) determines the execution order by considering priority, dependencies, parallelism, etc., and triggers Excel API calls. In the case of asynchronous execution, the processing flow is managed by a Promise or callback mechanism.

[0245] Next, the execution status monitoring step can be performed. The execution control module (64) monitors the progress and completion status of each task in real time and calls an exception handling routine when an error occurs.

[0246] Finally, the result integration and reporting step can be performed. When all tasks are completed, the execution control module (64) aggregates the results and transmits the execution logs and performance data to the system integration management module (90). The result data is transmitted to the AI ​​response processing module (80) and reflected in an Excel sheet.

[0247] According to the execution control module (64) described above, the execution plan generated by the AI ​​is executed through systematic scheduling and control rather than simple calling, thereby preventing conflicts between commands, data corruption, and duplicate execution. Additionally, an interactive Excel-AI integration environment is implemented where the user's natural language commands are automatically executed and results are reflected in real time. Furthermore, since all execution processes are recorded as logs and metadata, execution history management and post-mortem analysis (audit) are possible. Moreover, Excel resources can be managed stably even when multiple commands are requested simultaneously through the execution queue and scheduling functions. Furthermore, continuous automation execution without abnormal termination is supported through error recovery and retry mechanisms. Additionally, status and result information generated during execution is transmitted to the AI ​​response processing module (80), allowing the AI ​​to generate a context-aware subsequent response.

[0248] The result reflection module (65) can substantially reflect the execution results of the AI ​​execution plan generation module (50) and the execution control module (64) on the user's Excel worksheet or TaskPane UI.

[0249] The result reflection module (65) may be located at the final stage of the device and updates the results of operations performed according to AI or user requests (e.g., inserting formulas, generating data, creating charts, changing formats, generating summaries, etc.) in a consistent manner within the user's Excel environment.

[0250] Additionally, the result reflection module (65) can reflect the result value to the corresponding sheet location or visualization element while maintaining data linkage between the reference range, the basic work area, and the TaskPane UI.

[0251] The main functions of the result reflection module (65) may include a work result mapping and location determination function, a result format conversion and data reflection function, a UI update and visual feedback provision function, a change history recording and revert support function, and a result verification linkage function.

[0252] The operation result mapping and location determination function reflects the results generated within the AI ​​or device to target cells, sheets, and object locations according to the result mapping table of the reference history conversion module (33) and the execution control module (64). Here, the generated results encompass the entirety of execution result data (values, codes, objects) produced within the AI ​​execution flow, which ultimately becomes the target for actual reflection into cells, sheets, charts, etc., within the Excel environment by the result reflection module (65). For example, the generated results may include data operation results (sums, averages, statistical summaries, etc.) generated by the AI ​​analysis / response processing module (81) and / or the execution control module (64), formatting or structure change results (table creation, column addition, cell merging, etc.) generated by the execution control module (64), and text / code generation results generated by the AI ​​analysis module (42) and / or the AI ​​judgment module (43). The result mapping table has a structure that is initially generated by the reference history conversion module (33) and updated and corrected by the execution control module (64). That is, the reference history conversion module (33) converts a reference range (cell, table, chart, etc.) selected by the user into an internal reference ID to create an "input-output" mapping structure, and the execution control module (64) can dynamically modify where the actual execution result should be reflected (e.g., when creating a new sheet) during the execution process. The result mapping table is managed cooperatively by the two modules, but the entity responsible for its creation is the reference history conversion module (33).

[0253] The result format conversion and data reflection function automatically converts and inserts / updates results to fit Excel's data structure when they are generated in the form of text, formulas, tables, charts, images, or analysis reports.

[0254] The UI update and visual feedback features display real-time summary information of result reflection status, reflected cell ranges, or changed objects through the TaskPane UI.

[0255] The change history recording and undo support features save the previous state of the reflected results in the form of logs and metadata, allowing users to revert to the state prior to the change if desired.

[0256] The result verification linkage function links with the pre-execution verification module (63) and the exception handling module (71) to automatically check the consistency and validity of data before and after the result is reflected.

[0257] The operation flow of the above result reflection module (65) may be as follows.

[0258] First, a result reception step can be performed. The result reflection module (65) receives work execution result data from the execution control module (64) or the exception handling module (160). Here, the work execution result data may refer to the final result (e.g., data, formula, code, object, state information, etc.) produced after the execution procedure planned by the AI ​​execution plan generation module (50) is executed through the execution control module (64) or the exception handling module (71). For example, the work execution result data may include operation result data (value, array, statistical summary, etc.) and format / structure change result (object state data) generated by the execution control module (64), code / formula generation result (Formula, VBA Script, SQL syntax, etc.) generated by the AI ​​analysis / response processing module (81) or the execution control module (64), and error / exception handling result (Error Object, Log, Recovery Flag, etc.) generated by the exception handling module (71).

[0259] Next, result mapping and conversion steps can be performed. The result reflection module (65) maps the work result data to corresponding worksheet locations by referring to the cell reference information and reference mapping table provided by the reference history conversion module (33), and converts it according to the data format.

[0260] Next, a result reflection step can be performed. The result reflection module (65) reflects the converted result into an Excel cell, shape, chart, or data table. If necessary, the data structure after result reflection can be re-analyzed to perform automatic formatting alignment and visual feedback.

[0261] Finally, a feedback and recording step can be performed. The result reflection module (65) visually displays the reflected result on the TaskPane UI and saves the metadata of the task (result value, reflection time, target cell, etc.) as a log.

[0262] According to the result reflection module (65) described above, the user can automatically reflect the AI's calculation results in an Excel sheet without separate manual editing, thereby improving work speed and accuracy. Additionally, through integration with the reference history conversion module (33), the results are accurately reflected in corresponding cells and objects, maintaining logical consistency and visual consistency of the Excel document. Furthermore, through result visualization and status display via the TaskPane UI, the user can immediately check the work results and continuously perform additional commands if necessary. Moreover, the reflection results are stored as logs and metadata, making it easy to revert to a previous state in case of an error or to track and verify the results. Additionally, through a reflection structure linked with the pre-execution verification module (63) and the exception handling module (71), only error-free results are reflected, thereby strengthening the reliability of the entire AI automation process.

[0263] In FIG. 2, the exception and flow restoration processing module (70) detects exceptions occurring during execution, records error messages, determines the cause of the error through AI analysis, and supports re-execution or restoration according to the user's request. Here, execution refers to the entire operation in which the execution control module (64) actually performs the execution scenario defined by the AI ​​execution plan generation module (50), and the subject of the execution is the execution control module (64). The exception and restoration processing module (70) monitors all errors and exceptions occurring during this execution process and performs restoration or re-execution when necessary to maintain the stability of the entire flow. That is, execution may refer to the execution phase in which the execution control module (64) controls the execution scenario defined by the AI ​​execution plan generation module (50) and actually reflects it in the Excel environment (or its internal object). The execution targets (i.e., targets of exception monitoring) performed by the execution control module (64) may include data operations (sum, average, sort, filtering), insertion of formulas / functions, manipulation of objects (creation of tables, addition of charts, movement of shapes), and execution of macros / VBA (script-based automation).

[0264] The operation flow of the exception and flow restoration processing module (70) may be as follows.

[0265] First, the exception and flow restoration processing module (70) detects the error when an error occurs during execution. For example, the exception and flow restoration processing module (70) detects error codes, exception logs, abnormal delays, etc. by monitoring the execution status of the execution control module (64) in real-time based on events.

[0266] Next, the exception and flow restoration processing module (70) classifies the type of error (formula error, VBA error, missing reference, etc.). For example, the exception and flow restoration processing module (70) transmits error messages and log data to the AI ​​analysis module (42) or the internal LLM engine to classify the cause as a formula error, VBA error, missing reference, etc., and stores the classification result in the form of a log in the DB of the server (20).

[0267] Next, the exception and flow restoration processing module (70) analyzes the cause of the error through the AI ​​analysis module (42) and suggests generating a re-execution plan if the user desires.

[0268] Next, the exception and flow restoration processing module (70) restores the previous state based on the execution log when the user clicks the revert button or requests execution restoration.

[0269] Afterwards, the exception and flow restoration processing module (70) utilizes a backup copy of the data from the state prior to execution when performing restoration processing.

[0270] The exception and flow restoration processing module (70) may include an exception processing module (71) and an execution flow control and restoration module (72).

[0271] The exception handling module (71) detects input errors, data inconsistencies, execution failures, or abnormal user requests that may occur during the AI-based automated processing process and performs a function to maintain the stability of the system in response to such errors.

[0272] The exception handling module (71) ensures continuous operation of the Excel work environment by automatically generating an exception recovery path or inducing user intervention when execution is interrupted or an error occurs.

[0273] The exception handling module (71) supports the user's retry and correction process by intuitively displaying the cause of the error and recovery suggestions through interaction with the user interface module (31).

[0274] The main functions of the exception handling module (71) may include error detection and classification functions, automatic recovery or alternative path generation functions, user feedback interface provision functions, and log and diagnostic data recording functions.

[0275] The error detection and classification function detects errors occurring in the execution control module (64) or the AI ​​execution plan generation module (50) in real time and classifies them according to the type of error (e.g., syntax error, reference error, inability to perform operation, AI response failure, etc.).

[0276] The automatic recovery or alternative path generation function automatically generates an alternative execution path or performs a retry procedure for detected errors through predefined recovery rules or AI-based recommendations.

[0277] The user feedback interface provision feature supports users in immediately understanding the cause of errors by visually displaying error messages, cause analysis, and correction suggestions on the TaskPane UI described below.

[0278] The log and diagnostic data recording function stores the history of exceptions and responses in the form of logs and metadata, utilizing them as supporting data for preventing similar errors and improving AI models in the future.

[0279] The operation flow of the above exception handling module (71) may be as follows.

[0280] First, an error detection step can be performed. After receiving the result of the operation from the execution control module (64), the exception handling module (71) analyzes the result data or status code to determine whether there is an error.

[0281] Next, error classification and diagnosis steps can be performed. The exception handling module (71) classifies the detected errors by cause type (input error, missing reference, AI response failure, etc.) and evaluates whether the error is recoverable.

[0282] Next, a recovery attempt or user notification step can be performed. If the error is recoverable, the exception handling module (71) automatically executes an alternative execution plan based on internal rules or AI inference. If the error is unrecoverable, the user interface module (31) displays the cause of the error and a correction suggestion.

[0283] Finally, logging and reporting steps can be performed. The exception handling module (71) logs all exception occurrences and handling history and provides a summary report to an administrator or user when necessary.

[0284] According to the exception handling module (71) described above, immediate detection and response to errors are possible, thereby significantly improving the stability of the entire workflow. Additionally, operations can proceed continuously without manual user intervention through automatic recovery or alternative execution paths. Furthermore, the cause of the error and the solution are clearly presented on the UI, allowing the user to resolve the problem without a separate debugging process. Moreover, by accumulating error logs and recovery data, the prediction accuracy and resilience of the AI ​​execution plan generation module (50) can be continuously improved in the future. Additionally, consistent processing logic is applied to the same type of exception, preventing issues with different error responses depending on the user or situation.

[0285] The execution flow control and restoration module (72) manages the entire execution procedure of the device of the present invention in sequence units and controls the operation state to be restored or restarted when an abnormal situation such as an error / break / interrupt occurs.

[0286] The execution flow control and restoration module (72) is linked with the AI ​​execution plan generation module (50), execution control module (64), exception handling module (71), result reflection module (65), etc., to track and control the progress status, branch path, exception occurrence location, and restoreable state snapshot information of the AI's task execution flow in real time.

[0287] The execution flow control and restoration module (72) can manage the entire process of executing a complex data processing command through multiple modules sequentially and step-by-step, even if the user inputs the command in natural language within the TaskPane environment in Excel, and can stably restore and re-execute the command even if an error occurs in an intermediate stage.

[0288] The main functions of the execution flow control and restoration module (72) may include execution flow tracking and state management functions, task restoration and re-execution functions, branch flow management functions, execution history log and snapshot storage functions, and system stability and transaction consistency maintenance functions.

[0289] The execution flow tracking and state management functions manage status information for each processing stage of each module (e.g., request history classification module (41), AI execution plan generation module (50), pre-execution verification module (63), execution control module (64), result reflection module (65)) in the form of a flow tree or sequence table. Additionally, the completion status of each execution stage, the time of exception occurrence, and restoration points are stored as metadata.

[0290] The task restoration and re-execution function restores the execution state immediately prior to the interruption (input parameters, reference cells, execution scripts, result buffers, etc.) when an error or interruption signal is received from the exception handling module (71), and performs re-execution based on user approval or automatic judgment.

[0291] The branch flow management function is linked with the work request execution processing branch module (61) to independently manage the state of each branched flow when multiple work requests exist simultaneously or when the execution plan is configured in multiple stages (conditional).

[0292] The execution history log and snapshot saving features store step-by-step logs of the entire execution flow, result reflection points, error events, and restoration success status on a timeline basis, and allow users to restore the execution state at a specific point in time in snapshot units if desired.

[0293] The system stability and transaction consistency maintenance function manages data exchange between modules on a transaction basis to prevent data inconsistencies or duplicate reflections during interruptions or restarts.

[0294] The operation flow of the above execution flow control and restoration module (72) may be as follows.

[0295] First, the execution status registration step can be performed. The execution flow control and restoration module (72) collects the execution time and input data of each step from the execution control module (64) or the AI ​​execution plan generation module (50) and registers the status information in the execution flow management table.

[0296] Next, a real-time flow tracking step can be performed. The execution flow control and restoration module (72) updates the execution status whenever each module (e.g., request history classification module (41), AI execution plan generation module (50), pre-execution verification module (63), execution control module (64), result reflection module (65)) reports a processing completion or error status, and creates a branch path if necessary.

[0297] Next, an error detection and restoration preparation step can be performed. When the execution flow control and restoration module (72) receives an error event from the exception handling module (71), it retrieves the last normal execution snapshot prior to the point of error occurrence to obtain restoration target information (input value, execution parameter, reference cell, result buffer, etc.).

[0298] Next, restoration and re-execution steps can be performed. The execution flow control and restoration module (72) resumes execution from the corresponding restoration point when the user issues a retry command or when the system satisfies the automatic restoration condition. If the same error is repeated during the re-execution process, the error cause data is transmitted to the AI ​​execution plan generation module (50) to enable the establishment of a modified execution plan.

[0299] Finally, the result reflection and log recording steps can be performed. When normal restoration and re-execution are completed, the execution flow control and restoration module (72) updates the results on the sheet through the result reflection module (65) and records logs such as restoration history, processing time, and success status.

[0300] According to the execution flow control and restoration module (72) described above, even in complex natural language-based execution flows, immediate restoration and re-execution are possible in the event of an error or interruption, thereby minimizing system downtime. Additionally, since the input values, AI plans, and result buffers at the time of execution interruption are maintained, automatic resumption from the interruption point is possible without the need to repeatedly input the same command. Furthermore, by tracking the state of each execution stage, problems such as data inconsistency between modules, duplicate execution, and loss of results are prevented. Moreover, even when multiple job requests exist or conditional branches occur, the progress status of each flow is managed independently, maintaining consistency of the entire process. Additionally, execution flow logs and snapshot data are accumulated and can be utilized for subsequent verification and error analysis, and can also contribute to self-learning feedback for the AI.

[0301] In FIG. 2, the AI ​​response processing module (80) can determine whether a natural language command is an informational query based on the AI ​​execution judgment result of the request analysis and classification module (40). If the natural language command is an informational query, the AI ​​response processing module (80) can generate and output a natural language response sentence.

[0302] In other words, the AI ​​response processing module (80) can analyze a user request using AI and generate and output a corresponding response. The AI ​​response processing module (80) can automatically generate and provide a natural language response suitable for the user's request intent based on the analysis results derived from the AI ​​analysis module (42) and the AI ​​judgment module (43).

[0303] The AI ​​response processing module (80) can receive text or voice input from a user and perform response type classification, response content generation, conversation context maintenance and log recording, server-side integrated storage, etc. If necessary, the AI ​​response processing module (80) can also manage the context state for subsequent requests (e.g., "Explain in detail," "Create a table").

[0304] The main functions of the AI ​​response processing module (80) may include a request type determination function, a response generation and configuration function, a conversation log merging and summary storage function, a voice and text input processing function, and a response result delivery and display function.

[0305] The request type determination function classifies whether the input natural language command is an information request type (What, How, Why, etc.) or a work instruction type (execution, calculation, editing, etc.). By referring to the conversational / execution type distinction signal among the judgment results of the AI ​​judgment module (43), if it is determined to be conversational, it branches to the response generation routine of this module. If necessary, the context of the query and the previous conversation history (Log History) are analyzed together to identify whether it is a subsequent query or an independent query.

[0306] The response generation and configuration function configures the answer content through a path separate from the AI ​​execution plan generation module (50), based on key concept, target object, and relation information among the analysis results of the AI ​​analysis module (42). It automatically generates natural language sentences that match the user's request intent using a pre-trained language model (LLM) or a pre-defined rule-based response template. For readability and expression suitability, the generated response adjusts the sentence structure, notation, units, etc. according to post-processing rules, such as "The average value is 35.2" or "The total sum of column A is 1,240". When the response is linked to Excel data, it includes a cell identifier or reference result within the response using cell range information provided by the reference history conversion module (33).

[0307] The conversation log merging and summary saving function combines user input (text / voice) and AI responses into a single conversation unit (Dialog Turn) to generate Dialog Log Data. The generated logs, including time information, user ID, command type, response summary, and reference information for used data, are recorded in an integrated storage area within the server database. Repetitive query and response patterns within the conversation logs are summarized and can be utilized as AI training data or for generating future personalized recommendations.

[0308] In the voice and text input processing function, user voice input is converted into text via a Speech-to-Text (STT) conversion engine and fed into the same natural language analysis pipeline. Responses can be provided not only in text form but also converted via Text-to-Speech (TTS) to serve as voice feedback. This enables users to perform voice-based conversational AI operations even within the Excel environment.

[0309] The response result delivery and display function displays the generated response through the TaskPane UI or popup interface window of the user interface module (31). The response may automatically include an action button for additional operations (e.g., "execute the corresponding calculation," "visualize with a chart"). If necessary, the response content may be delivered to the execution control module (64) and used as input for subsequent command execution.

[0310] The operation flow of the above AI response processing module (80) may be as follows.

[0311] First, an input reception step can be performed. The AI ​​response processing module (80) receives the user's text or voice input and preprocesses it into a form that can be recorded in the conversation log.

[0312] Next, a request determination step can be performed. The AI ​​response processing module (80) determines that the current user's request (command) is an informational (descriptive) query based on the result of the AI ​​judgment module (43).

[0313] Next, a response planning step can be performed. The AI ​​response processing module (80) determines the response type (description, summary, definition, procedure guide, etc.) using the request context and analysis results, and performs relevant data referencing if necessary.

[0314] Next, a natural language response generation step can be performed. The AI ​​response processing module (80) uses a language model (LLM) to generate a response in the form of a contextually appropriate natural language sentence. If necessary, formula results or table data may be included in the form of a string. For example, if the user's request (command) is an informational (descriptive) query (e.g., "Summarize this table," "How does it compare to last month?", "Explain what role this function plays"), a natural language response sentence is generated.

[0315] Next, a response display step can be performed. The AI ​​response processing module (80) outputs a response to a display area (TaskPane, Chat Area, etc.) within the user interface module (31). If necessary, a follow-up request button or execution option may be presented together.

[0316] Finally, a log storage and summary step can be performed. The AI ​​response processing module (80) stores the query-response pairs of this conversation session in the server DB and records a summary together. Subsequently, when the same user reconnects, the previous conversation context can be restored.

[0317] According to the AI ​​response processing module (80) described above, by immediately understanding the user's natural language commands and providing natural responses, even non-experts can intuitively perform Excel automation. Additionally, through the merging and summarizing functions of conversation logs, the context is not interrupted, and responses can be provided by continuing the previous state exactly as it was when subsequent questions or requests for modification are made. Furthermore, unnecessary re-analysis is reduced, and rapid responses can be generated based on recent conversation history. Additionally, since it supports both text and voice, its usability is expanded in various work environments (meetings, presentations, etc.). Moreover, all conversation history is recorded on the server (20) side and can be used for subsequent analysis, verification, quality control, and auditing. Furthermore, the response logs can be refined and summarized to be used as data for AI retraining, thereby enabling the system's self-improvement.

[0318] The AI ​​response processing module (80) may include an AI analysis / response processing module (81) and an AI conversation management module (82).

[0319] The AI ​​analysis / response processing module (81) analyzes the user's natural language request, extracts data ranges, cell reference information, command types, etc. related to the request intent, converts the request into a specific AI execution plan or query (SQL, formula, API command, etc.), and processes the result so that it is output in a natural response form to the user interface (UI).

[0320] The AI ​​analysis / response processing module (81) may internally include a pre-trained language model, a rule-based engine, or a user-customized fine-tuning model, and may perform advanced query interpretation and response generation by connecting with an external server (20; cloud server).

[0321] The main functions of the AI ​​analysis / response processing module (81) may include natural language query analysis, cell reference / data mapping, execution plan conversion support, response generation and summary, learning / adaptive response, and error detection and automatic correction suggestion functions.

[0322] The natural language query analysis function breaks down user input (e.g., "Show me the total sales for this month") into syntactic and semantic units and classifies request types (query, summary, calculation, visualization, etc.).

[0323] The cell reference / data mapping function analyzes reference information (e.g., A1:D10, Sheet2!B5, etc.) transmitted from the cell reference selection detection module (32) or the reference history conversion module (33) and maps the data table and field names associated with the query.

[0324] The execution plan conversion support function transmits the analyzed query intent and mapping results to the AI ​​execution plan generation module (50) to support the configuration of executable workflow steps (e.g., data loading → filtering → calculation → visualization).

[0325] The response generation and summary function generates a natural language-based response (e.g., "Total sales for this month are 25,480,000 won") for the execution results or analyzed data and sends it to the user interface module (31).

[0326] The learning / adaptive response function receives the context of the same user or similar query from the AI ​​conversation management module (82) and generates a consistent response that fits the context or automatically corrects the terminology.

[0327] The error detection and automatic correction suggestion function detects syntax errors, reference errors, etc. that occur during query analysis or execution, and in conjunction with the exception handling module (71), suggests an automatic correction or a user confirmation request.

[0328] The operation flow of the above AI analysis / response processing module (81) may be as follows.

[0329] First, an input reception step can be performed. The AI ​​analysis / response processing module (81) receives natural language input text, reference cell information, and request tags (summary, graph generation, etc.) from the user interface module (41).

[0330] Next, an intent interpretation step can be performed. The AI ​​analysis / response processing module (81) separates the natural language sentence into morphemes and semantic units, and the AI ​​model or rule engine infers the purpose of the request (e.g., aggregation, comparison, filtering). At this time, the AI ​​analysis / response processing module (81) improves interpretation accuracy by referencing the previous conversation context (previous query, filter conditions, etc.) from the AI ​​conversation management module (82).

[0331] Next, a data reference mapping step can be performed. The AI ​​analysis / response processing module (81) analyzes the reference structure transmitted from the cell reference selection detection module (32) and the reference history conversion module (33) and matches it with the data field or range required by the query.

[0332] Next, the execution plan delivery step can be performed. The AI ​​analysis / response processing module (81) transmits the interpreted intent and mapped data structure to the AI ​​execution plan generation module (50) to generate an actual executable processing step.

[0333] Next, a result response generation step can be performed. When the AI ​​analysis / response processing module (81) receives result data from the AI ​​execution plan generation module (50) or the execution control module (64), it converts it into natural language and presents it intuitively to the user. If necessary, the AI ​​analysis / response processing module (81) can simultaneously generate visualization results (charts, summary tables, etc.) and transmit them to the result reflection module (65).

[0334] Finally, a feedback / learning step can be performed. The AI ​​analysis / response processing module (81) recognizes user feedback ("Summarize it again," "Compare it to last month") and updates the weights of the response model or stores them in context memory. Here, context memory is one of the core components of the AI ​​analysis / response processing module (81) and may refer to an internal memory structure for temporarily or continuously storing and managing the conversation context (context) between the user and the AI, interaction history, and the state of the analyzed data reference. In other words, context memory may be defined as a conversation state storage unit within the AI ​​analysis / response processing module (81) that stores the intent, reference context, result context, and recent response trace accumulated during the conversation process with the user. Unlike a general memory cache, context memory may include the ability to store, update, and delete semantic context information rather than simply storing data.The context memory can store items such as user intent data (Intent Data; for example, the result of natural language interpretation of the purpose, action, object, etc. that the user intended to perform in a recent command), reference data (Contextual Reference; for example, information on object references within Excel such as cell references, sheet names, data table ranges, etc.), execution result history (Result Trace; for example, the result of executing a previous command, the result of generating a summary, analysis table, chart, etc.), dialogue state information (Dialog State; for example, the progress status of the current conversation (summarizing, comparing, generating statistics, etc.)), and user feedback log (User Feedback; for example, supplementary requests made by the user ("explain it again," "show it in a different format"), etc.) while the AI ​​analysis / response processing module (81) is operating.

[0335] According to the AI ​​analysis / response processing module (81) described above, even non-experts can perform complex operations and analysis using only natural language queries without directly manipulating Excel or databases. Additionally, by linking with the AI ​​conversation management module (82) to maintain the context of continuous queries, a conversational analysis environment can be implemented. Furthermore, the AI ​​automatically detects and corrects reference range errors and syntax inconsistencies, thereby minimizing user intervention. Moreover, query interpretation and execution plan formulation are automated, reducing execution preparation time and improving system processing efficiency. Additionally, analysis results are provided in natural language and visual elements (tables, charts) for intuitive understanding, and are immediately reflected in a worksheet in conjunction with the result reflection module. Furthermore, by linking with an external server (cloud server), large-scale data analysis and improved response quality using advanced language models are possible.

[0336] The AI ​​conversation management module (82) manages the flow of natural language conversation between the user and the AI, maintains context, and enables continuous understanding and response to multiple requests.

[0337] The AI ​​conversation management module (82) may be located between the user interface module (31) and the AI ​​execution plan generation module (50). The AI ​​conversation management module (82) manages natural language requests entered by the user on a conversation session basis and controls the provision of appropriate responses and execution suggestions by understanding the context.

[0338] The AI ​​conversation management module (82) can continuously accumulate user queries and system responses to improve the accuracy of AI responses and suggest user-customized execution flows (e.g., recommending follow-up commands, providing error guidance, assisting with result interpretation, etc.).

[0339] The main functions of the AI ​​conversation management module (82) may include conversation session management, conversation context (Cintext) tracking, intent and response type classification, conversation-based execution control, AI response generation and delivery management, and conversation log and learning feedback functions.

[0340] The conversation session management function creates and maintains conversation sessions for each user and stores the request and response history for each session. It maintains the conversation flow to contextually connect and process consecutive user requests (e.g., "Calculate the sum of this column → Show me the average as well") within the same session.

[0341] The conversation context tracing function reflects contextual information when interpreting the meaning of a new request by referencing the target cell, calculation range, and function executed of the previous command. It recognizes previous response results or error messages and reflects them in subsequent conversations.

[0342] The intent and response type classification function analyzes the type of request transmitted from the request history classification module (41) to distinguish whether the request is a Q&A, execution, or help / guide type. Based on the classified type, an appropriate response strategy (e.g., establishing an execution plan, providing knowledge, or interpreting errors) is selected.

[0343] The conversation-based execution control function transmits additional commands or follow-up questions from the user to the execution control module (64) and the AI ​​execution plan generation module (50) when they affect the existing execution flow, thereby triggering adjustment or expansion of the execution process.

[0344] The AI ​​response generation and delivery management function refines the response returned from the AI ​​model or external language engine to fit the display format within the Excel UI (e.g., message window, tooltip, cell description, highlight, etc.). In cooperation with the result reflection module (65), it reflects user feedback in real time.

[0345] The conversation log and learning feedback function stores all conversation content (requests, responses, error messages, follow-up queries, etc.) as a structured log and transmits it to the server (20) in the form of metadata so that it can be used for the training and performance improvement of the AI ​​model in the future.

[0346] The operation flow of the above AI conversation management module (82) may be as follows.

[0347] First, the session creation and initialization steps can be performed. When a user enters a natural language command in the Excel UI (TaskPane or conversation input window), the user interface module (31) transmits it to the AI ​​conversation management module (82), and if the user's conversation session does not exist, a new one is created.

[0348] Next, context recognition and request analysis steps can be performed. The AI ​​conversation management module (82) determines whether the current request is a continuation of an existing task or a new task by referring to the previous conversation history and cell information of the reference history conversion module (33).

[0349] Next, an intent and response type classification step can be performed. The AI ​​conversation management module (82) determines whether the response is a simple question response, an execution trigger, or an explanation-assisted type based on the request type (e.g., calculation, filtering, explanation, error query) transmitted from the request history classification module (41).

[0350] Next, the steps of generating an AI response and linking execution can be performed. If the response is of the execution type, the AI ​​conversation management module (82) transmits a command to the AI ​​execution plan generation module (50) to establish an execution plan. If the response is of the query type, the AI ​​conversation management module (82) generates a response from an internal knowledge base or an external AI model.

[0351] Next, the steps of returning a response and displaying the UI can be performed. The AI ​​conversation management module (82) transmits the generated response to the user interface module (31). Accordingly, the generated response is displayed in the form of a conversation window, a notification message, or a cell comment. The user can enter additional commands or provide feedback on the result.

[0352] Finally, the history saving and context updating steps can be performed. The AI ​​conversation management module (82) saves all request and response contents as session history and reuses them as context information in the next conversation.

[0353] According to the AI ​​conversation management module (82) described above, the user can naturally perform a series of commands such as "sort this result → calculate the average → show it as a chart," and the system can continue the AI ​​execution flow by remembering the context of the previous command. In addition, since Excel tasks can be performed using only natural language conversation without knowledge of complex commands or functions, even non-expert users can efficiently perform data analysis and report automation. Furthermore, by continuously accumulating conversation history and feedback information, the AI ​​model can learn the user's preferences and work patterns. Additionally, since the AI ​​conversation management module (82) is linked in real-time with the user interface module (31), execution control module (64), and result reflection module (65), complex controls such as execution, suspension, modification, and restart are possible even during a conversation. Furthermore, by transmitting information such as conversation logs, response accuracy, and user feedback to the server (20), continuous performance improvement and quality management of the AI ​​model are possible.

[0354] The system integration management module (90) serves as a central control hub for the device of the present invention and can comprehensively monitor and manage data flow, execution status, resource usage, and error and exception handling status between each module.

[0355] In addition, the system integration management module (90) coordinates communication between internal modules (e.g., AI execution plan generation module (50), execution control module (64), result reflection module (65), etc.) and external linkage modules (e.g., cloud AI server, external DB, API linkage interface, etc.) and performs the role of maintaining processing consistency, resilience, security, and execution efficiency of the entire system.

[0356] That is, the system integration management module (90) can be seen as performing integrated coordination, verification, logging, and version control so that individual function modules operate independently but data and control flows do not conflict with each other.

[0357] The main functions of the above-mentioned system integration management module (90) may include a module integration status management function, a data flow control and path management function, a resource and session management function, a log and history management function, an error recovery and process resumption function, an external server / AI cloud integration management function, a version and configuration management function, and a policy and permission integration management function.

[0358] The module linkage status management function periodically monitors the execution status (waiting, running, completed, error) of each module (reference history conversion module (33), AI analysis module (42), execution control module (64), exception and flow restoration processing module (70), result reflection module (65), system integration management module (90), etc.) and updates the integrated status table when the status changes.

[0359] Data flow control and path management functions manage the paths for data transmission between modules (e.g., reference information, execution results, analysis responses, etc.) and prevent unnecessary duplicate transmissions or collisions.

[0360] The resource and session management function comprehensively monitors user session, memory, CPU, and network usage, and adjusts resource allocation priorities.

[0361] The log and history management function ensures traceability by integrating and storing execution logs, error history, and data change history of all modules.

[0362] The error recovery and process resumption function works in conjunction with the execution flow control and restoration module (72) to specify a restore point and resume the process when a system interruption or error occurs.

[0363] The external server / AI cloud integration management function manages communication sessions with cloud AI servers, external databases, API engines, etc., and controls authentication keys, transmission speeds, and network status in real time.

[0364] The version and configuration management function maintains compatibility within the system by managing module-specific build versions, schema versions, and AI model versions.

[0365] The integrated policy and permission management function enhances security by centrally managing user-specific access rights, executable module scope, and data access control.

[0366] The operation flow of the above system integration management module (90) may be as follows.

[0367] First, a module status monitoring step can be performed. The system integration management module (90) periodically collects the status of each module (execution status, response delay, exception occurrence, etc.). The collected status information is managed as a Module State Map to identify the overall operation flow in real time.

[0368] Next, the execution request integration and routing steps can be performed. When the initial request is input from the user interface module (31), the execution branch request is transmitted to the system integration management module (90) via the request history classification module (41). The system integration management module (90) routes the request to the AI ​​execution plan generation module (50), the execution control module (64), or the AI ​​analysis / response processing module (81) depending on the type of request.

[0369] Next, the data transmission and flow management steps can be performed. The system integration management module (90) tracks and manages data exchanged between modules (e.g., analysis results, reference history, execution logs, etc.) and controls packet exchange and transmission queues to maintain a synchronized state.

[0370] Next, error detection and restoration control steps can be performed. When the system integration management module (90) receives an error signal from the exception handling module (71) and the execution flow control and restoration module (72), it performs a retry or re-run according to the automatic restoration policy.

[0371] Next, the result integration and reporting step can be performed. The system integration management module (90) aggregates the results received from the result reflection module (65), records them in a log, and reflects them in the user's session or cloud synchronization storage.

[0372] Finally, external integration and resource optimization steps can be performed. When AI computation using an external server (cloud server) (20) or an external API is included, the system integration management module (90) optimizes the resource usage, bandwidth, and authentication status of the session in real time.

[0373] According to the system integration management module (90) described above, the execution status of each module can be managed in an integrated manner to prevent conflicts between modules, duplicate execution, and data loss. Additionally, through organic linkage with the exception handling module (71) and the execution flow control and restoration module (72), automatic restart or reprocessing can be performed in the event of an error. Furthermore, by centrally adjusting execution priority and resource allocation, overall processing delays are minimized and high-performance processing is possible. Moreover, information leakage is prevented by managing user-specific execution rights, AI request scope, and data access levels in an integrated manner. Furthermore, various components such as external AI servers, databases, and plugin modules can be easily integrated. Additionally, logs, history, and version information are unified to improve the efficiency of system maintenance and fault analysis. Furthermore, in conjunction with the pre-execution verification module (63), system-level quality standards (data integrity, result reliability, etc.) are automatically checked.

[0374] FIG. 3 is a flowchart of natural language-based AI automation execution according to an embodiment of the present invention.

[0375] Referring to FIG. 3, the present invention operates in a Microsoft Excel (100) environment, and when a user selects an object such as a cell, shape, picture, chart, etc. in the Excel basic workspace (110) and inputs a command in natural language, an artificial intelligence module can analyze it and perform appropriate automation execution.

[0376] In FIG. 3, Microsoft Excel (100) is an application environment in which a user performs tasks, and includes a basic Excel workspace (110) in which various objects such as cells, shapes, pictures, and charts are placed, and a TaskPane UI (120) which is an AI interaction interface.

[0377] The Excel basic workspace (110) allows the user to select specific cells, shapes, pictures, or charts within Excel, and the selected objects are subsequently provided as input data for AI analysis.

[0378] The cell reference selection detection module (32) detects the user's selection actions in real time. That is, the user can select desired objects such as cells, shapes, pictures, and charts in the Excel main workspace (110), and this selection information is detected in real time through the cell reference selection detection module (32).

[0379] The cell reference selection detection module (32) performs detailed detection functions of text selection detection (112), shape and picture selection detection (113), and chart selection detection (114) depending on the type of selected object. This detection result is provided as context information for the AI ​​analysis step.

[0380] TaskPane UI #2 (120) is a primary interaction interface that allows a user to input requests in the form of natural language, and performs functions such as command input, execution path selection, and execution confirmation through a chat interface (123). TaskPane UI #2 (120) can be described as a natural language command input interface.

[0381] TaskPane UI #2 (120) chat interface ("Type message here") (123) receives natural language input and questions.

[0382] The conversation display area (121, 122) within TaskPane UI #2 (120) is a conversation window in which questions and answers are displayed in a thread structure.

[0383] The XLFormula / VBA buttons in TaskPane UI #2 (120) are user interfaces (UI) that provide options for converting and applying the output format of the response result. For example, when the XLFormula button is selected, the response is provided as an Excel formula, and the formula can be inserted into the current selection area by clicking the Apply button. When the VBA button is selected, macro code is generated, and the user can select to register, sign, or execute after reviewing the code. Therefore, the XLFormula button or the VBA button should be understood as trigger buttons that allow the response to be converted and applied as an Excel formula or VBA code when the result is already presented as a response on the user screen.

[0384] In the AIExecData generation (130) step, the request history classification module (41) first classifies the user's natural language input into response requests, information requests, and work requests.

[0385] When a task request is detected, the task request execution processing branch module (61) guides the user to the execution path options suggested by the AI ​​(e.g., Excel formula-based processing, VBA automation-based processing, or both methods). At this time, the system explains the advantages and disadvantages of each and presents the options via a chat bubble button within the TaskPane so that the user can decide the execution path themselves.

[0386] Next, the task execution target selection module (62) determines whether to reflect the result in a new sheet or based on a specific cell range, depending on the execution path selected by the user.

[0387] The AI ​​execution plan generation module (50) creates specific execution steps according to the selected execution path and structures the AIExecData object in units of Excel formulas or VBA code.

[0388] The execution flow control and restoration module (72) finalizes the user's execution selection, records the execution restoration structure, and provides an execution / return button.

[0389] Finally, in the request execution (150) step, an automation task within the Excel environment is executed based on the confirmed AIExecData.

[0390] As described above, when a user selects a specific object (cell, chart, etc.) in the Excel environment and inputs a natural language command, the cell reference selection detection module (32) analyzes the selection information, the user interface module (31) receives the command, the request history classification module (41) and the task request execution processing branch module (61) determine the execution path, the task execution target selection module (62) specifies the reflection location, the AI ​​execution plan generation module (50) generates a formula / VBA execution plan, the execution flow control and restoration module (72) controls whether to execute, and the execution control module (64) performs automation in the actual Excel environment. Through this, the user can perform intuitive and accurate automation in Excel based on natural language requests without separate complex procedures, and the system guarantees reproducibility, safety, and resilience for the entire execution flow.

[0391] In other words, through the execution path-based flow described above, users can combine objects selected in the Excel environment with natural language requests, directly select the desired method from among various execution options presented by the AI, and reliably and quickly perform tasks such as formula creation, VBA execution, and data processing.

[0392] FIG. 4 is a flowchart of an Excel cell or object reference selection branch according to an embodiment of the present invention.

[0393] Referring to FIG. 4, the device of the present invention automatically detects object information such as cells, shapes, pictures, and charts selected by the user in a Microsoft Excel (210) environment and performs a reference information conversion procedure to reflect this in an AI analysis request.

[0394] The Excel basic workspace (212) includes a general Excel document (213) that the user is editing, and the user can select a cell (214), a shape (213), or a chart (214). If a chart is selected, the series reference area (215) of the chart is also recognized.

[0395] Selected object information is classified by type through a cell reference selection detection module (32). The cell reference selection detection module (32) has four branching paths: no selection, text selection detection, shape and picture selection detection, and chart selection detection.

[0396] Each selection type is structured and processed by the reference history conversion module (33). For example, in the case of a cell selection, the cell selection history conversion JSON module (231) converts information such as the address, value, and format of the selected cell into JSON form. In the case of an image object (shape or picture), the image JSON conversion module (232) includes the image data by encoding it in Base64 format. In the case of a chart selection, the chart information JSON conversion module (233) encodes the chart image in Base64, and the reference range JSON conversion module converts the chart's SERIES formula and cell reference range into JSON data.

[0397] In other words, the cell selection history conversion JSON module (231) can convert the address, value, format, merge status, formula, etc. of a cell or cell range selected by the user in the Excel workspace into Reference Object JSON in a JSON format suitable for the internal data model. Looking at the operation flow of the cell selection history conversion JSON module (231), when location information (sheet name, row / column coordinates, cell range) of the cell selected by the user is received from the user interface module (31) or the cell reference selection detection module (32), the type of data contained in each cell (number, string, date, etc.) is identified, and whether a formula is included and formatting attributes (font, background color, border, etc.) are extracted. Then, the extracted information is mapped into a standardized key value structure (e.g., "sheet", "range", "value", "formula", "style") and converted into a JSON object. The converted JSON object (or data) is transmitted to the AI ​​analysis module (42) or the task execution target selection module (62).

[0398] The image JSON conversion module (232) can analyze the location, size, identifier, link, and tag meta information of unstructured visual objects, such as pictures, shapes, and icons inserted into an Excel sheet, and convert them into reference data in JSON format. Looking at the operation flow of the image JSON conversion module (232), it obtains a list of attributes of all image-type objects existing in the current sheet through the Excel API or Object Model, and extracts identification ID, insertion sheet name, coordinates (X, Y), size (Width, Height), link path, alternative text, and layer order information for each object. The extracted information is mapped into a JSON format and structured into fields such as "objectType":"image", "position", "size", and "meta". If necessary, Base64 encoded data of the image itself is attached, or a server storage path is referenced. The generated JSON object is utilized by the AI ​​analysis module (42) for analysis or report generation tasks that include visual elements.

[0399] The chart information JSON conversion module (233) analyzes the data source range, chart type, axis information, legend, labels, visualization attributes, etc., of a chart (Chart Object) created within Excel and converts them into JSON format so that they can be utilized for AI analysis or automatic report generation. Looking at the operation flow of the chart information JSON conversion module (233), it extracts chart identifiers, data source cell ranges, chart types (e.g., bar, line, pie), axis and legend information from the Excel Chart Object, and maps data point values, colors, display formats, and legend items for each series. The extracted information is converted into a JSON object containing fields such as "chartType", "dataRange", "series", "axis", "legend", and "style". The JSON data is directly used for chart updating, data interpretation, and comparative visualization tasks in the AI ​​analysis module (42) or the AI ​​response processing module (80).

[0400] The JSON data converted in this way is combined with the user natural language request history (222) and transmitted to the natural language request and reference history AI analysis request (234).

[0401] Subsequently, based on the data, the AI ​​analyzes the object context associated with the user's command intent and generates a suitable automation execution plan.

[0402] Accordingly, as shown in Fig. 4, by automatically detecting and structuring object information selected by the user within Excel, it is possible to accurately provide context data necessary for AI analysis, improve the accuracy of execution of automation commands without separate manual specification, and minimize user input steps.

[0403] FIG. 5 is a diagram showing the process of branching the utterance history, branching the work request execution method, verifying the execution plan, and branching the execution according to an embodiment of the present invention.

[0404] When a user inputs a natural language request (300) in an Excel environment, the AI ​​response processing module (80) of the device of the present invention analyzes the meaning of the utterance and classifies it into one of the types of a conversation request (310), an information request (320), or a task request (330).

[0405] First, when classified as a conversation request (310), the AI ​​response processing module (80) performs the chat window response and termination (311) step, thereby terminating the process with a conversational response with the user.

[0406] Next, in the case of an information request (320), the AI ​​analysis module (42) performs an information search plan establishment (321) step to determine the source and search method of the necessary data. Subsequently, the device of the present invention provides an output method selection (322) step so that the user can specify how to display the results, at which time one of creating a new sheet or outputting based on a specific cell can be selected.

[0407] When the information search is completed, the analysis results and search process are presented to the user in the result explanation (323) step, and the user may approve or reject the actual execution in the execution selection (324) step. If execution is selected, the process proceeds to the common processing (342) step, and if not executed, the flow ends in the termination (341) step.

[0408] Meanwhile, in the case of a task request (330), the device of the present invention first performs a step of determining the processing method (331) according to the nature of the requested task to determine whether processing based on Excel formulas or VBA automation scripts is possible. It also determines whether the generation of output results is necessary. If output is required, it provides a step of selecting the output method (332) so that the user can specify the location to display the results, and then explains the detailed procedure and expected results of the task in the step of explaining the execution plan (333). The user specifies whether to actually execute through the step of selecting whether to execute (334), and if execution is approved, it proceeds to the common processing (342) step, and if not executed, it ends in the termination (341) step.

[0409] In the common processing (342) step, a series of processes are performed, including merging conversation records on the server (20) side, updating existing conversation summary data, and creating an AI execution schema. This AI execution schema is subsequently used for automation execution within Excel, so that the search for requested information or the execution of tasks is automatically performed on the actual worksheet.

[0410] According to Figure 5, the present invention enables efficient and safe AI-based automation tasks within Excel by automatically selecting an appropriate processing path based on the user's utterance and supporting the user to clearly determine whether to execute it.

[0411] In other words, it provides a procedural structure that enables AI automation tasks within Excel to be performed in a more stable and user-controllable manner by naturally branching the processing flow based on the user's utterance and explicitly confirming whether to execute.

[0412] Figure 6 is a diagram showing the detailed processing flow for the information request step illustrated in Figure 5.

[0413] When a user inputs a natural language request in an Excel environment and it is classified as an information request (320) through utterance type analysis, the device of the present invention first performs the information search plan establishment (321) step.

[0414] In other words, when a user inputs a request composed of natural language in an Excel environment, the device of the present invention analyzes the utterance and classifies it as an information request (320), and can perform processing according to the following steps.

[0415] First, the device of the present invention performs an information search plan establishment step (321) in response to an information request utterance. That is, when it is confirmed that the user's natural language utterance is for the purpose of information search, the AI ​​analysis / response processing module (81) performs the information search plan establishment step (321). In this step, the AI ​​constructs a search path that matches the intent and purpose of the request, establishes a search plan including the expected search purpose and search method, result output format, and related data structuring method, and displays the contents on the UI through the user interface module (31).

[0416] Afterward, the user can determine the output format of the result in the output method selection step (322) after verifying the search plan. The output method can be selected from outputting to a new sheet or outputting based on a specific cell. This selection is subsequently used to determine the location where the data is reflected. In other words, in the output method selection step (322), the user selects whether to output the result to a new sheet or to a specific cell. This user selection is detected and reflected by the task execution target selection module (62), and subsequently influences the location where the result is reflected during AI processing.

[0417] Subsequently, the selected output method, user request content, and Excel reference data are transmitted together in the information request AI analysis transmission (323) step. For example, the selected output method, the natural language content of the user request, and reference information including cell or object information (shapes, charts, etc.) selected by the user in Excel are transmitted to the AI ​​analysis module (42) through the information request AI analysis transmission step (323). In other words, when the user's natural language request is classified as an information request, the request history classification module (41) identifies the type of the request, and the reference history conversion module (33) structures the Excel reference data (i.e., information on the cell or object (shapes, charts, etc.) selected by the user) into JSON format. This structured data is transmitted to the AI ​​analysis module (42) in the information request AI analysis transmission step (323).

[0418] In the AI ​​exploration and data processing execution (324) stage, the AI ​​performs the necessary data exploration and processing tasks according to the transmitted request and generates information description and output components. During this process, common processing (342) is performed simultaneously, merging the existing conversation history on the server side with the new request / response history and updating the conversation history summary. This is always performed regardless of whether a final output is produced.

[0419] In other words, the AI ​​analysis / response processing module (81) performs necessary information retrieval, refinement, and analysis based on the received request and reference data during the AI ​​search and data processing execution step (324). As a result, it generates summarized information descriptions and output components (tables, charts, numbers, etc.). Additionally, common processing (342) is performed in parallel during this step, which is necessary for conversation consistency and history management. That is, in parallel with the AI ​​response processing, the AI ​​conversation management module (82) performs the common processing step (342). The AI ​​conversation management module (82) executes functions such as merging existing conversation logs with current request / response content on the server (20) side, updating conversation summary data in real time, and backing up and logging conversation history. Common processing (342) is always performed in parallel regardless of whether the user executes it, and contributes to conversation-based history management and execution tracking of the device of the present invention.

[0420] The AI ​​search results are summarized and provided to the user in the result explanation (325) step, where they can be viewed in the form of data output (326) or data preview (327). If the user does not select execution, the flow ends with non-execution termination (341). In other words, the search results generated by the AI ​​are summarized and provided to the user in the result explanation step (325). On the user interface, the user can view or select results in a manner such as data output (326), which means immediate output, and data preview (327), which means performing only result prediction. All of these interface operations are performed through the user interface module (31).

[0421] If the user does not select execution, the processing flow ends at the non-execution termination step (341).

[0422] On the other hand, if the user selects execution, the process proceeds to the output execution (329) stage, and the execution plan is generated as an AI execution schema and reflected in the Excel environment. In other words, when execution is confirmed, the AI ​​execution plan generation module (50) automatically generates an AI execution schema (AIExecData) based on the request content, reference data, and selected output format, and the execution control module (64) automatically performs output operations within the Excel environment based on the generated execution schema. The executed result is reflected (data values, formulas, visualization elements, etc.) in a sheet or cell by the result reflection module (65).

[0423] According to FIG. 6, the present invention enables a request history classification module (41) to identify the request type for a user's natural language utterance for the purpose of information inquiry in an Excel environment, a reference history conversion module (33) to structure selection information, an AI analysis / response processing module (81) to establish a search plan and generate response data, and to receive confirmation from the user regarding output and execution through a user interface module (31). When execution is confirmed, an AI execution plan generation module (50), an execution control module (64), and a result reflection module (65) perform automation step by step, and an AI conversation management module (82) can merge and record all conversation and execution logs. By doing so, the entire process from information inquiry to result reflection is automated into a structured flow, and a natural language-based AI-Excel linkage system designed to enable user-centric intuitive control is implemented.

[0424] That is, according to FIG. 6, the present invention systematically configures the selection of an output method, AI-based data search and processing, common data management processing, and execution determination for an information request utterance, thereby enabling the user to quickly and accurately retrieve and utilize necessary information in an Excel environment.

[0425] Figure 7 is a diagram showing the detailed processing flow for the work request step illustrated in Figure 5.

[0426] When a user inputs a natural language request in an Excel environment and it is classified as a task request (330) through utterance type analysis, the device of the present invention first performs a step of determining a processing method (331). In this step, the AI ​​determines whether the requested task can be processed based on Excel formulas or based on VBA automation, and if both methods are possible, it first suggests the formula-based method. In addition, if it is a VBA method, it also determines whether output (code generation) is required.

[0427] In other words, when a user instructs a task in the form of natural language in an Excel environment, the device of the present invention determines the utterance type of the request and classifies it as a task request (330). When the user's natural language request is classified as a task type in this way, the task request execution processing branch module (61) performs a processing method determination step (331). In this step, the AI ​​automatically determines whether the task can be processed based on Excel formulas or based on VBA automation, suggests the formula-based method first if both methods are possible, and determines whether code output is necessary in the case of the VBA method. This determination result is provided to the user in the form of a processing method suggestion and is directly reflected in the establishment of future execution plans and the selection of execution logic.

[0428] Subsequently, in the processing method selection step (332), the user selects either an Excel formula or a VBA method and specifies whether to apply the result to a new sheet or based on a specific cell. The selected processing method, user request details, and Excel reference data are transmitted together in the task request AI analysis transmission (335) step. In other words, the user selects one of the methods suggested by the AI ​​in the processing method selection step (332). The selectable items are the processing method (Excel formula method or VBA code method) and the output location (new sheet or based on a specific cell). The user's selection is detected and stored by the task execution target selection module (62) and is included together in the subsequent AI analysis request.

[0429] The user's selection details and reference data are transmitted to the AI ​​processing module through the task request AI analysis transmission step (335). The information transmitted together at this time includes the selected processing method and output format, the natural language content of the user request, and Excel reference data (structured information such as selected cells, charts, shapes, etc.). In other words, the user's selection information and request details are transmitted to the request history classification module (41) and the reference history conversion module (33) in the task request AI analysis transmission step (335). Here, the request history classification module (41) analyzes the utterance content to reconfirm that it is a task request type, and the reference history conversion module (33) converts the information such as Excel cells, charts, and shapes selected by the user into structured JSON data. The transmitted data includes the natural language request, selection method, and reference data.

[0430] In the AI ​​processing plan establishment (336) stage, the AI ​​creates (generates) an AI execution plan that includes the requested processing procedure, expected result, and exception handling method, and displays the plan in the user chat window. During this process, common processing (342) is performed simultaneously to merge the existing conversation history on the server side with the new request / response history and update the conversation history summary. This always proceeds regardless of whether the final execution is performed. In other words, the transmitted data is analyzed by the AI ​​execution plan generation module (50), and an execution plan is created (established) that includes the procedure and logic of the processing task requested in the AI ​​processing plan establishment stage (336), the expected result, and the exception handling method (e.g., cell protection, reference error, security restriction, etc.). The generated execution plan is output to the user through the chat interface. And, simultaneously with the AI ​​processing plan establishment, the AI ​​conversation management module (82) performs the common processing stage (342). The AI ​​conversation management module (82) maintains the state consistency of the entire system by merging server-side conversation logs with new requests / responses, updating conversation history summaries, backing up execution history, and managing records by user. These common processing (342) are always processed in parallel regardless of whether the user executes them.

[0431] In the case of formula-based processing, the result suitability verification step (337) checks whether the expected result of the formula generated by the AI ​​matches the user's requirements and re-establishes the processing plan if necessary. In other words, in the result suitability verification step (337), the pre-execution verification module (63) pre-determines whether the formula automatically generated by the AI ​​meets the user's requirements and, if necessary, modifies the execution plan or suggests an alternative.

[0432] In the case of VBA-based processing, the validation step (338) checks for compilation errors in the code, and if errors are found, establishes a correction plan. In other words, in the validation step (338), the pre-execution verification module (63) examines whether the generated VBA code is executable without compilation errors, identifies the location and details of the error when an error occurs, and establishes a correction plan or provides a user notification.

[0433] Afterwards, in the execution selection step (334), the user can check the processing history summary and select either the execution statement preview (327) or the immediate execution (326).

[0434] When execution (326) is selected, an AI execution schema (329) is generated and the corresponding operation is performed in the Excel environment. If the user does not wish to execute, the flow is terminated with non-execution termination (341). In other words, when the user selects execution, the AI ​​execution plan generation module (50) generates an AI execution schema (AIExecData) based on the AI ​​execution plan, the execution control module (64) performs operations such as formula execution, VBA code execution, and data output, and the result reflection module (65) reflects the execution results (formulas, values, charts, etc.) to a designated Excel location.

[0435] On the other hand, if the user does not select execution, the processing flow ends at the non-execution termination step (341).

[0436] According to FIG. 7, the present invention implements a natural language-based AI-Excel interaction system that ensures reliability, verifiability, and user control over the automated processing of Excel tasks.

[0437] That is, according to FIG. 7, the present invention systematically configures the determination of processability, selection of method, establishment of AI-based execution plan, prior verification, common data management processing, and determination of execution for a work request utterance, thereby enabling a user to safely and efficiently perform automation tasks in an Excel environment.

[0438] FIG. 8 is a diagram showing the process of execution flow control and AI execution schema generation according to an embodiment of the present invention.

[0439] When a task requested by a user in the Excel environment is transmitted to the execution request reception (410) stage, the system performs the AI ​​execution schema generation (420) stage. In this stage, an execution schema is generated that includes the user request content, Excel reference data, the location where the result is reflected, and execution log items. In other words, among the tasks requested by a user in the Excel environment in natural language form, requests requiring execution are transmitted to the system through the execution request reception stage (410). These requests are received through the user interface module (31), and the request history classification module (41) classifies them as executable requests among the task / information types. Then, the AI ​​execution plan generation module (50) performs the AI ​​execution schema generation stage (420) based on the received requests. In the AI ​​execution schema generation stage (420), an AI execution schema (AIExecData structure) is configured that includes items such as the user request content, Excel reference data, the location where the execution result is reflected, execution log items, and meta information. The generated AI execution schema is transmitted to the execution control module (64) in the subsequent stage and is used as reference information to control execution.

[0440] Subsequently, in the pre-execution verification (430) step, the sheet protection status is checked, the target cell lock and PIVOT TABLE status are examined, and in the case of VBA execution, the macro security setting status is checked. If the verification is passed, the execution control (440) step is entered. In other words, the pre-execution verification module (63) performs the pre-execution verification step (430) to conduct a preliminary check on whether execution is possible. The preliminary check items at this time include whether the target Excel sheet is protected (checking the sheet protection status), whether the target cell is locked and whether a PIVOT TABLE is included, and, if the request is VBA-based, checking the VBA macro security setting status. Such preliminary checks (verification) prevent errors during execution in advance and ensure that execution suitable for the user's environment conditions is performed.

[0441] Execution control (440) is further divided into sub-procedures of data output (441), formula execution (442), and VBA execution (443). In data output (441), a new sheet is created or values ​​are output to designated cells, and in formula execution (442), formulas are placed at designated locations and sheet recalculations are performed. In VBA execution (443), the created VBModule is bound and VB functions are called to check the execution status. In other words, the execution control module (64) performs actual Excel operations through the execution control stage (440) and consists of sub-procedures (detailed execution procedures) such as data output (441), formula execution (442), and VBA execution (443). In the case of data output (441), new sheets are created or values ​​from designated cells are output, and output formatting (table format, style reflection, etc.) is applied. In the case of formula execution (442), formulas according to user requests are placed at designated cell locations, and formula recalculations are performed for the entire sheet or a partial range. In the case of VBA execution (443), the generated VBModule is bound, VBA functions are called through macro execution, and the execution status is checked. Each of these detailed procedures (sub-procedures) is branched according to the processing method of the requested task and is flexibly configured to encompass all formula / automation / VBA code executions.

[0442] If an error occurs during execution, the process moves to the exception handling (460) stage to record the error log, analyze the type of error, and request the AI ​​to review the cause. If necessary, the user's request for re-execution is checked, and upon approval, the process re-enters the execution control (440) stage. In other words, if an error occurs during execution, the process switches to the exception handling stage (460). The exception handling module (71) performs error log generation and storage, error type analysis and error message classification, and error cause review through a request for AI analysis. If necessary, the exception handling module (71) may also request the user to confirm whether to re-execute. If, as a result of the error analysis, re-execution is approved through user confirmation, the flow re-enters the execution control (440) stage to proceed with recovery execution.

[0443] When the execution is successfully completed, the result is applied to a sheet or cell in the result reflection (450) step. In other words, when the execution is successfully completed, the result reflection module (65) performs the result reflection step (450). At this time, the result of the formula execution, the result of VBA processing, etc., is reflected in a designated cell or a new sheet, and the result including numerical data, charts, images, etc., is also applied. The reflected result is immediately displayed on the user's Excel work screen, allowing the user to intuitively check the results of the automation processing.

[0444] Subsequently, in the final termination (470) stage, the execution log is saved on the server side, the server conversation record is updated, and the entire process is completed. In other words, when the entire execution flow is completed, the system performs the final termination stage (470). In the final termination stage (470), the execution log is saved on the server (20) side, the user's conversation record is merged and updated on the server (20), and the restorable execution flow data is managed. In the final termination stage (470), the execution flow control and restoration module (72), the system integration management module (90), and the AI ​​conversation management module (82) are responsible for ensuring data consistency and backing up logs.

[0445] According to FIG. 8, based on the AI ​​execution plan schema, the user interface module (31) and the request history classification module (41) receive execution requests, the AI ​​execution plan generation module (50) structures the execution schema, the pre-execution verification module (63) checks the environment status, the execution control module (64) is responsible for data output, formula execution, and VBA execution, the exception handling module (71) performs a reprocessing routine when an error occurs, the result reflection module (65) reflects the final result in the Excel sheet, and the system integration management module (90) and the AI ​​conversation management module (82) store the termination log and conversation record on the server (20), thereby ensuring the reliability, recoverability, and systematic nature of execution history management of natural language-based Excel automation execution.

[0446] That is, according to FIG. 8, the present invention enables safe and reliable AI-based automation in an Excel environment by systematically organizing a series of procedures based on an AI execution schema, including pre-execution verification, granular execution control, error analysis and re-execution flow, result reflection, and log management.

[0447] The foregoing description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains will be able to make various modifications and variations within the scope of the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are intended to explain, not limit, the technical concept of the present invention, and the scope of the technical concept of the present invention is not limited by these embodiments. The scope of protection of the present invention shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0448] 10: Natural language-based Excel automation support device 20: Server 30: User Interaction Interface Module 40: Request Analysis and Classification Module 50: AI Execution Plan Generation Module 60: Task Execution Flow Control Module 70: Exception and Flow Restoration Handling Module 80: AI Response Processing Module 90: System Integration Management Module

Claims

Claim 1 A natural language-based Excel automation task support device operating in a Microsoft Excel environment, comprising: a user interaction interface module that receives natural language commands from a user, detects a reference object selected upon input of a natural language command, converts the detected reference object into an internal reference structure, and outputs it as user request data; a request analysis and classification module that receives the user request data, analyzes the intent, action, and context of the natural language command, converts the analysis results into structured command interpretation information, determines a corresponding execution type based on the analysis results, and outputs it as an AI execution judgment result; an AI execution plan generation module that receives the AI ​​execution judgment result and concretizes the natural language command into actual execution procedure units to generate an AI execution schema; a task execution flow control module that receives the AI ​​execution schema, selects a target object for task execution, calls an internal execution routine corresponding to the determined execution type to perform the actual task, and outputs the actual task result as task execution result data; and an AI response processing module that determines whether the natural language command is an informational query based on the AI ​​execution judgment result of the request analysis and classification module, and if the natural language command is an informational query, generates and outputs a natural language response sentence.A natural language-based Excel automation task support device comprising: a request analysis and classification module that analyzes the natural language command to identify the intent and reference target of the natural language command, and classifies the natural language command into one of the types of information request, conversation request, and task request based on the identified result; if the natural language command is classified as a task request, the task execution flow control module determines whether the task can be processed based on Excel formulas or should be processed based on VBA automation and displays the result on the UI through the user interaction interface module; the AI ​​execution plan generation module receives information including the processing method selected and input by the user, the output format, the natural language content of the user request, and reference data; the AI ​​execution plan generation module analyzes the received information to generate an AI execution plan and displays it on the UI through the user interaction interface module; and, depending on the user's execution selection, the AI ​​execution plan generation module generates an AI execution schema based on the AI ​​execution plan, and the task execution flow control module executes the task based on the AI ​​execution schema. Claim 2 delete Claim 3 A natural language-based Excel automation task support device according to claim 1, wherein if the natural language command is classified as a conversation request, the AI ​​response processing module performs a termination step after a chat window response. Claim 4 A natural language-based Excel automation task support device according to claim 1, wherein when the natural language command is classified as an information request, the AI ​​response processing module constructs a search path that matches the intent and purpose of the request, establishes a search plan including an expected search purpose and search method, a result output format, and a related data structuring method, and displays the established search plan on a UI through the user interaction interface module. Claim 5 In claim 4, when the user interaction interface module receives a confirmation event of a search plan displayed on the UI and receives a selected output method, the content of a user request, and reference data, the user interaction interface module transmits the selected output method, the content of a user request, and the reference data to the request analysis and classification module; the request analysis and classification module identifies the type of the user request; the user interaction interface module structures the reference data and transmits it to the request analysis and classification module; the AI ​​response processing module performs information retrieval, refinement, and analysis based on the user request and reference data to generate summarized information descriptions and output components and provide them to the user; and as the user interaction interface module receives the user's execution selection event, the AI ​​execution plan generation module generates an AI execution schema based on the content of the user request, reference data, and the selected output format; and the task execution flow control module performs an output operation within an Excel environment based on the AI ​​execution schema, a natural language-based Excel automation task support device. Claim 6 delete Claim 7 A natural language-based Excel automation task support device according to claim 1, wherein the AI ​​execution judgment result includes the intent of the natural language command, the action of the natural language command, the context, and the execution type. Claim 8 A natural language-based Excel automation task support device, further comprising: an exception and flow restoration processing module that detects an error or exception occurring during the execution of the task execution flow control module, classifies the error type and saves it as a log, determines whether restoration is possible, and performs re-execution or restoration according to a user request. Claim 9 A natural language-based Excel automation task support device, further comprising: a system integration management module that integrates and manages result data, conversation logs, and execution logs from the AI ​​response processing module, stores execution logs on an external server, and updates conversation records and user-specific execution history in the first paragraph.

Citation Information

Patent Citations

  • Self-learning based dialogue apparatus for incremental dialogue knowledge, and method thereof

    KR1020170088164A

  • Apparatus and method for processing natural language query about relational database using transformer neural network

    KR1020230081594A

  • System for managing Robotic Process Automation System and Driving method thereof

    KR1020230100260A

  • Generating structured results using machine learning

    KR1020250121554A