User-Attributed Agent Proxy Action and Permission-Controlled Interaction System, Method, and Recording Medium Using a Multimodal Evidence-Based Persona Graph

The user-attributed agent system with a multimodal evidence-based persona graph addresses the limitations of existing technologies by securely managing and controlling persona information and actions, ensuring reliable and privacy-compliant interactions.

KR103044468B1Active Publication Date: 2026-11-06LINKBRICKS HORIZON-AI INC
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
KR1020260076355
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2026-04-28
Publication Date
2026-11-06
Estimated Expiration
2046-04-28

AI Technical Summary

Technical Problem

Existing technologies lack a comprehensive structure for managing persona information reliability, controlling delegated actions, ensuring safety and accountability, differentiating access scopes based on disclosure levels, and protecting personal information in multi-agent interactions.

Method used

A user-attributed agent system using a multimodal evidence-based persona graph that extracts and manages persona attributes from various data sources, controls interactions based on authority, and performs proxy actions securely while ensuring privacy and accountability.

Benefits of technology

Enhances the reliability and explainability of persona-based interactions, maintains user representativeness, reduces privacy risks, and enables safe and reliable proxy actions by managing persona attributes and their sources, reliability, and usage permissions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026051514693-PAT00001_ABST
    Figure 112026051514693-PAT00001_ABST
Patent Text Reader

Abstract

The present invention relates to a system, a method, and a recording medium for performing the same, which generates a persona evidence graph linking persona attributes with the source, reliability, visual information, disclosure level, and usage permission scope of corresponding grounds from multimodal user data, configures a user-attributed agent based thereon, and controls it to perform various interactions and proxy actions within the scope of authority. Furthermore, the present invention provides the effect of simultaneously ensuring privacy protection, secure execution of surrogate actions, authority-based information utilization, and post-event traceability by integrally performing authority-based transformation of the persona evidence graph, policy and risk-based control of surrogate actions by behavior type, control of interactions between multiple agents, and audit records for the entire process.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a system and method for proxy behavior and authority control interactions of a user-attributed agent using a multimodal evidence-based persona graph, and a recording medium for performing the same. More specifically, the invention relates to a system and method for proxy behavior and authority control interactions of a user-attributed agent using a multimodal evidence-based persona graph, and a recording medium for performing the same, wherein the system extracts individual persona attributes from multimodal user data including documents, images, conversation history, user input, and behavior records, generates a persona evidence graph including the source, reliability, disclosure level, and usage permission scope of the evidence corresponding to the persona attributes, and wherein the user-attributed agent performs a response or action on behalf of the user based on the persona evidence graph, and controls the interaction between multiple user-attributed agents within the scope of authority. Background Technology

[0002] With the recent rapid proliferation of conversational AI systems based on Large Language Models, technologies related to personalized conversation, personal assistant functions, automated responses, multi-agent collaboration, user simulation, and digital personas are developing rapidly.

[0003] These technologies are evolving beyond single-model-centric response generation structures to utilize multiple AI agents to perform more sophisticated interactions and problem-solving.

[0004] For example, Korean Registered Patent Publication No. 10-2051384 (registered on November 27, 2019) discloses a multi-agent system comprising a target language analysis agent, a self-emotion identification agent, a conversational situation recognition agent, a decision-making agent, and a response generation agent, and proposes a structure that recognizes the situation and generates a response based on the semantic and emotional information of the conversation.

[0005] In addition, U.S. Patent Publication US 11,431,660 B1 (registered August 30, 2022) discloses a collaborative conversational system in which multiple conversational artificial intelligence agents collaboratively generate responses through mutual message exchange.

[0006] However, the above-mentioned conventional technologies have the following limitations.

[0007] First, even if individual personas are extracted and updated over the long term from documents, images, conversations, and behavioral records, there are limitations in terms of the reliability of persona information and control over its use due to the lack of a structure that manages the basis from which each persona attribute was derived, as well as the source, reliability, scope of disclosure, and scope of permission for use of that basis.

[0008] Second, even if an agent exists to act on behalf of the user, there is a lack of a structure to systematically manage the scope of delegated actions, the requirement for approval, controls based on behavioral risk, and the possibility of auditing and tracing the results of external system integration, which limits the ability to ensure safety and accountability in actual service environments.

[0009] Third, even when multiple user-associated agents participate in interactions such as discussions or negotiations, there is insufficient structure to finely differentiate the scope of each agent's access to other users' information based on disclosure levels or usage permissions, and to enable interactions based on persona information reconstructed at different levels.

[0010] Fourth, when providing conversations similar to those with specific third parties, the structure for controlling the scope of use by distinguishing between permitted and unpermitted data and restrictively configuring persona information according to permission conditions is insufficient, making it difficult to meet the requirements for personal information protection and permission-based data utilization.

[0011] Fifth, while technology exists for personalized AI to generate user-customized responses, a structure is not sufficiently established to sequentially perform actual external actions in the absence of a user, and to record and manage the basis, approval process, and permitted scope of such actions so that they can be verified retrospectively.

[0012] Therefore, although the aforementioned prior art technologies partially provide multi-agent-based interaction and response generation functions, they had limitations in comprehensively implementing evidence-based management, authorization control, proxy action execution, multi-party interaction, and auditing capabilities for personas generated from multimodal user data. The problem to be solved

[0013] The present invention has been devised to solve the problems of the prior art described above. The objective of the present invention is to provide a user-attributed agent generation structure based on a persona evidence graph that extracts persona attributes from multimodal user data including documents, images, conversation history, user input, and behavior records, and stores the source, reliability, visual information, disclosure rating, and usage permission scope of the evidence corresponding to the persona attributes in conjunction.

[0014] In addition, another objective of the present invention is to enable the user-attributed agent to perform proxy actions, including external API calls or workflow execution, rather than merely generating a response in the absence of a user or under pre-delegation conditions, while ensuring that such actions are securely controlled based on behavioral risk and scope of authority.

[0015] Furthermore, another objective of the present invention is to enable authority-based information utilization and interaction control by using persona evidence that is transformed differently according to the relationship, disclosure level, and usage scope between the requesting subject and the user identified by the user persistent identifier, when multiple user-attributed agents perform discussions, negotiations, or conversations representing the perspectives of different individuals.

[0016] In addition, another objective of the present invention is to provide an approval policy structure that selects the proxy action as one of an action that can be automatically performed, an action that requires user approval, or a blocking action, depending on the policy and risk level of the action type.

[0017] In addition, another objective of the present invention is to store the persona attributes, rationale, transformation results, approval history, and execution logs used throughout the entire process of the interaction and proxy behavior as audit records, thereby enabling subsequent tracking and verification of which persona attributes and rationale were used in which responses or actions. means of solving the problem

[0018] To achieve the above objectives, the system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph according to the present invention comprises: an input collection unit that collects multimodal user data including documents, images, conversation history, user input, or behavior records associated with a user; a preprocessing unit that preprocesses the multimodal user data into a form suitable for attribute extraction; an attribute extraction unit that extracts persona attribute candidates and attribute evidence metadata corresponding to the persona attribute candidates from the preprocessed multimodal user data; a persona graph generation unit that generates a persona evidence graph by linking the persona attribute candidates with a user persistent identifier, source, trust level, visual information, disclosure level, and usage permission scope; and an authority conversion unit that generates converted persona evidence by converting the persona attributes or evidence included in the persona evidence graph according to the relationship between the requesting subject and the user identified by the user persistent identifier, the disclosure level, the usage permission scope, and behavioral risk level. The system is characterized by comprising: an agent configuration unit that creates or updates a user-attributed agent based on the persona evidence graph or the transformed persona evidence; an interaction control unit that controls the user-attributed agent to operate in at least one interaction mode among a single proxy mode, a two-way conversation mode, a representative discussion mode, or a third-party persona conversation mode; a proxy action control unit that controls the user-attributed agent to perform at least one proxy action among message creation, schedule proposal, approval request creation, external API call, or workflow execution; and an audit log unit that stores persona attributes, grounds, transformation results, approval history, and execution logs used in the process of performing the interaction mode or the proxy action.

[0019] Here, the persona graph generation unit is characterized in that the persona evidence graph includes a user node, an attribute node, a basis node, and a relationship edge, wherein the basis node includes a source identifier, a time of creation, a reliability rating, a disclosure rating, and a license scope, and the relationship edge is configured to represent at least one of a relationship where the basis supports the attribute, a relationship where the basis conflicts with the attribute, a relationship where the attribute is derived from a parent attribute, a relationship that can be disclosed to a specific group of relationships, or a relationship where a specific license period expires.

[0020] Here, the authority conversion unit is characterized by being configured to generate the converted persona evidence by performing at least one of masking, de-identification, categorization, summarization, or removal of non-disclosure attributes on the grounds or attributes when the requesting entity cannot directly access the grounds or attributes included in the persona evidence graph.

[0021] Here, the proxy behavior control unit is characterized by being configured to classify the proxy behavior into one of an action that can be performed automatically, an action requiring user approval, or a blocking action based on a policy table by action type, generate an approval request if the risk level of the action is greater than or equal to a threshold value, and control the proxy behavior to be performed automatically if the risk level of the action is less than the threshold value and automatic execution is permitted by the policy table by action type.

[0022] Here, the proxy behavior control unit is characterized by being configured such that the behavioral risk is calculated based on at least one of data sensitivity, external system influence, monetary effect, legal effect, user absence status, or lack of reliability of grounds.

[0023] Here, the interaction control unit is characterized by being configured to verify a pre-registered permission package for a third party in the third-party persona conversation mode and to activate the third-party persona conversation mode only if at least one of the permission period, purpose of use, type of permitted data, and prohibited act included in the permission package is valid.

[0024] In addition, the interaction system for proxy behavior and authority control of a user-attributed agent using a multimodal basis-based persona graph according to the present invention further includes a moderation unit that evaluates the result of the interaction mode or the result of the proxy behavior to generate a final response or a final action result; the moderation unit is characterized by being configured to generate the final response or the final action result based on at least one of persona match, basis reliability, authority suitability, behavioral risk, issue coverage, or user constraint compliance. Effects of the invention

[0025] With the above configuration, the interaction system for proxy behavior and authority control of a user-attributed agent using a multimodal evidence-based persona graph according to the present invention has the advantage of improving the explainability and reliability of the basis upon which a generated response or performed action is derived by managing together the persona attributes extracted from multimodal user data and the source, reliability, disclosure level, and usage permission scope of the corresponding evidence.

[0026] In addition, depending on the relationship between the requesting subject and the representative user, the disclosure level, and the scope of use, even if the persona attributes are the same, persona evidence transformed in ways such as masking, anonymization, categorization, or summarization is utilized, which has the advantage of maintaining user representativeness while minimizing the exposure of personal information.

[0027] In addition, by controlling the selection of automatic execution, waiting for user approval, or blocking of proxy actions based on policies and risk levels for each type of action, there is an advantage in that safe and reliable proxy actions can be performed even in the absence of the user.

[0028] In addition, even when multiple user-assigned agents conduct discussions or negotiations representing different perspectives, each agent is configured to use only persona evidence within the permitted range transformed by the authorization conversion unit, which has the advantage of enabling interactions that reflect diverse perspectives while reducing the possibility of privacy infringement.

[0029] In addition, by performing data usage scope verification based on permission packages in the third-party persona conversation mode, allowed and disallowed data can be clearly distinguished, which has the advantage of preventing the misuse of third-party related information.

[0030] In addition, by integrating and storing the persona attributes, rationale, transformation history, approval details, and execution results used throughout the entire process of interaction and surrogate behavior in the audit log, there is an advantage in being able to track and verify the history of persona changes and the results of surrogate behavior performance retrospectively.

[0031] In addition, the structure described above offers the advantage of being applicable to various fields, such as customer support, collaborative decision-making, approval automation, education and mentoring, public decision support, and community management. Brief explanation of the drawing

[0032] FIG. 1 is a block diagram showing the overall configuration of a proxy behavior and authority control interaction system of a user-attributed agent using a multimodal basis-based persona graph according to an embodiment of the present invention. Figure 2 is a block diagram showing the processing flow for generating a persona evidence graph from multimodal user data. Figure 3 is a block diagram showing an authority conversion flow that converts persona evidence according to the authority scope of the requesting subject. Figure 4 is a block diagram showing the control flow for a proxy action that assesses the risk and determines whether to approve when a proxy action is performed by a user-attributed agent. Figure 5 is a block diagram illustrating the process of performing a discussion and deriving results in a representative discussion mode in which multiple user-associated agents participate. Figure 6 is a block diagram showing the flow of verifying permission packages and determining whether the conversation mode is active in third-party persona conversation mode. Figure 7 is a conceptual diagram showing the log structure and trace information stored in the audit log. FIG. 8 is a flowchart showing the processing flow of a method for surrogate behavior and authority control interaction of a user-attributed agent using a multimodal basis-based persona graph according to an embodiment of the present invention. Specific details for implementing the invention

[0033] Hereinafter, a system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal basis-based persona graph according to the present invention will be described with reference to the embodiments illustrated in the drawings.

[0034] FIG. 1 is an overall configuration block diagram of a proxy behavior and authority control interaction system of a user-attributed agent using a multimodal basis-based persona graph according to an embodiment of the present invention.

[0035] Referring to FIG. 1, a proxy behavior and authority control interaction system of a user-attributed agent using a multimodal basis-based persona graph according to an embodiment of the present invention is configured to include an input collection unit (10), a preprocessing unit (20), an attribute extraction unit (30), a persona graph generation unit (40), an authority conversion unit (50), an agent configuration unit (60), an interaction control unit (70), a proxy behavior control unit (80), a moderation unit (90), an audit log unit (100), a learning update unit (110), and a version management unit (120).

[0036] First, the input collection unit (10) will be explained.

[0037] The input collection unit (10) is configured to collect multimodal user data including documents, images, conversation history, user input, or behavior records associated with the user.

[0038] In addition, the input collection unit (10) may be configured such that the collected multimodal user data includes at least one of a text document, an image or a document containing an image, a messenger conversation, an email, a voice conversation transcript, a self-introduction input, a profile question response, a schedule information, a work history, a purchase history, an approval history, or external profile data.

[0039] At this time, the input collection unit (10) may be configured to include an input collection module (11), a session linkage module (12), and a user identification module (13).

[0040] The input collection module (11) can perform the function of collecting multimodal user data from various channels.

[0041] For example, data can be collected from messengers, emails, collaboration tools, schedule management systems, e-commerce systems, or external profile provision services, and can also include unstructured data such as image files or scanned documents.

[0042] The session linkage module (12) can perform the function of linking collected multimodal user data on a user session basis.

[0043] For example, conversation history, input logs, or behavior records of the same user at different points in time can be combined into a single continuous session to enable time-series-based analysis in a subsequent stage.

[0044] The user identification module (13) can perform user identification for the collected multimodal user data and perform the function of linking with the user persistent identifier.

[0045] For example, it can be configured to identify the same user based on login information, device identification information, account identifier, or external authentication information, and to consistently manage the user's data.

[0046] Additionally, the input collection unit (10) can be configured to record the source, creation time, collection time, and data type of the collected data together so that they can be utilized in the subsequent process of attribute extraction and basis metadata generation.

[0047] Accordingly, the input collection unit (10) collects multimodal user data in an integrated manner and manages it while maintaining consistency on a user-by-user basis, thereby performing the role of providing basic data that enables accurate extraction of persona attributes and evidence-based analysis in subsequent stages.

[0048] Next, the preprocessing unit (20) will be described.

[0049] The preprocessing unit (20) is configured to preprocess the multimodal user data into a form suitable for attribute extraction.

[0050] In addition, the preprocessing unit (20) may be configured to perform at least one of document optical character recognition (OCR), sentence separation, tokenization, named entity recognition, conversational speaker separation, image feature extraction, personal identification information detection, sensitive information tagging, time series sorting, or duplicate removal on the multimodal user data.

[0051] At this time, the preprocessing unit (20) may be configured to include an OCR module (21), a text normalization module (22), a named entity recognition module (23), a speaker separation module (24), and an image feature extraction module (25).

[0052] The OCR module (21) can perform the function of extracting text from an image or scanned document, and through this, can convert text information contained in unstructured image data into a structured form.

[0053] The text normalization module (22) can refine text data into a form suitable for analysis by performing sentence separation, tokenization, and removal of unnecessary symbols. In addition, it may include functions to standardize expressions of the same meaning or correct typographical errors.

[0054] The named entity recognition module (23) can perform the function of identifying and tagging named entities such as personal names, organization names, place names, dates, and amounts, and thereby can secure meaningful information units during the subsequent attribute extraction process.

[0055] The speaker separation module (24) can be configured to distinguish speakers for conversation data and identify which user generated each utterance.

[0056] For example, the accuracy of persona analysis can be improved by distinguishing user-specific utterance patterns in messenger conversations or meeting records.

[0057] The image feature extraction module (25) can perform the function of extracting visual features other than objects, scenes, or text from image data, and thereby support the extraction of image-based persona attributes.

[0058] In addition, the preprocessing unit (20) can identify personal information or sensitive information by detecting personal identification information and tagging sensitive information, and then provide it in the form of metadata so that it can be used in the authorization conversion unit (50).

[0059] Furthermore, the preprocessing unit (20) can be configured to perform time-series sorting and duplicate removal of collected data to sort the data flow for the same user in chronological order and prevent analysis errors caused by duplicate information.

[0060] Accordingly, the preprocessing unit (20) performs the role of providing a preprocessing foundation that integrates and refines various forms of multimodal user data, thereby enabling the attribute extraction unit (30) to generate accurate and consistent persona attribute candidates and basis metadata.

[0061] Meanwhile, FIG. 2 is a block diagram showing the overall processing flow in which a persona evidence graph is generated from multimodal user data by an input collection unit (10), a preprocessing unit (20), an attribute extraction unit (30), and a persona graph generation unit (40).

[0062] Referring to FIG. 2, multimodal user data collected by the input collection unit (10) is refined and structured through the preprocessing unit (20), the preprocessed data is extracted into persona attribute candidates and attribute basis metadata by the attribute extraction unit (30), and subsequently, the extracted information is generated into a persona evidence graph composed of user nodes, attribute nodes, and basis nodes by the persona graph generation unit (40).

[0063] Next, the attribute extraction unit (30) will be described.

[0064] The attribute extraction unit (30) is configured to extract persona attribute candidates and attribute basis metadata corresponding to the persona attribute candidates from the preprocessed multimodal user data.

[0065] In addition, the attribute extraction unit (30) may be configured to extract at least one of demographic attributes, occupation, professional field, interests, hobbies, value judgment criteria, preferences, dislikes, decision priority, speech characteristics, writing style characteristics, length characteristics, politeness characteristics, response patterns by relationship group, or repetitive behavior rules from the preprocessed multimodal user data as the persona attribute candidate, and to generate attribute basis metadata for each persona attribute candidate that includes an attribute name, attribute value, source of basis, reliability score, extraction time, and extraction method identifier.

[0066] At this time, the attribute extraction unit (30) may be configured to include an attribute candidate extraction module (31), a basis metadata generation module (32), and a reliability calculation module (33).

[0067] The attribute candidate extraction module (31) can perform the function of deriving various persona attribute candidates for the user by analyzing preprocessed text, image, and conversation data.

[0068] For example, interests repeatedly mentioned by a specific user, decision-making tendencies in specific situations, and characteristics such as tone of voice and politeness in conversation can be extracted and defined as candidate persona attributes.

[0069] The basis metadata generation module (32) can perform the function of generating basis information corresponding to each persona attribute candidate.

[0070] In this case, the supporting information may include the source, creation time, data type, and contextual information of the original data from which the relevant attribute was derived, and specific emails, messenger conversations, document content, or image analysis results may be linked as supporting evidence.

[0071] The reliability calculation module (33) can perform the function of calculating the reliability of the above attribute candidates and grounds.

[0072] For example, confidence scores can be calculated by assigning high confidence when the same attribute is repeatedly observed from various sources, or by assigning higher weight to more recent data.

[0073] In addition, the attribute extraction unit (30) can manage conflicting attribute candidates by maintaining them in parallel or assigning priorities, and can be configured so that they can be used to resolve conflicts in the persona graph generation unit (40) or version management unit (120).

[0074] Accordingly, the attribute extraction unit (30) systematically extracts various persona attributes representing user characteristics and corresponding grounds and reliability from multimodal user data, thereby providing key information for generating a persona evidence graph.

[0075] Next, the persona graph generation unit (40) will be explained.

[0076] The persona graph generation unit (40) is configured to generate a persona evidence graph by linking the persona attribute candidates with a user persistent identifier, source, trust level, visual information, public rating, and usage permission scope.

[0077] In addition, the persona graph generation unit (40) may be configured such that the persona evidence graph includes a user node, an attribute node, a basis node, and a relationship edge, and the basis node may be configured to include a source identifier, a time of creation, a reliability rating, a public rating, and a usage permission scope.

[0078] At this time, the above relationship edge may be configured to represent at least one of a relationship where the basis supports the attribute, a relationship where the basis conflicts with the attribute, a relationship where the attribute is derived from a parent attribute, a relationship that can be disclosed to a specific group of relationships, or a relationship where a specific usage period expires.

[0079] The persona graph generation unit (40) may be configured to include a graph node generation module (41), a relationship edge generation module (42), and a public rating / permission scope linkage module (43).

[0080] The graph node creation module (41) can perform the function of creating user nodes, attribute nodes, and basis nodes.

[0081] For example, interest attributes for a specific user and the supporting data for those interests can be created as attribute nodes and supporting nodes, respectively, and connected to user nodes.

[0082] The relationship edge generation module (42) can perform the function of generating an edge that defines the relationship between the nodes.

[0083] For example, a specific basis can establish a relationship in which a specific basis supports a specific attribute, a conflict relationship between conflicting attributes, or a relationship representing a hierarchical structure between attributes.

[0084] The public rating / permission scope linkage module (43) can perform the function of linking the public rating and permission scope for each node or relationship.

[0085] This allows for differential control of the range of information that can be exposed depending on the accessing entity, even if the persona attributes are the same.

[0086] In addition, the persona graph generation unit (40) can be configured to allow the generated persona evidence graph to be expanded over time, and can be configured so that when new attributes or grounds are added, updates are made while maintaining the relationship with the existing graph structure.

[0087] Accordingly, the persona graph generation unit (40) integrates attribute and basis information about the user into a structured graph form, thereby providing a data structure that can be efficiently utilized in the subsequent authority-based information utilization and agent creation stages.

[0088] Next, the authority conversion unit (50) will be explained.

[0089] The authority conversion unit (50) is configured to generate converted persona evidence by converting the persona attributes or grounds included in the persona evidence graph according to the relationship between the requesting subject and the user identified by the user persistent identifier, the public level, the usage permission scope, and the behavioral risk level.

[0090] In addition, the authority conversion unit (50) may be configured to generate the converted persona evidence by performing at least one of masking, de-identification, categorization, summarization, or removal of prohibited disclosure attributes on the grounds or attributes when the requesting entity cannot directly access the grounds or attributes included in the persona evidence graph.

[0091] At this time, the authority conversion unit (50) may be configured to include an access authority verification module (51), a data conversion module (52), and an authority metadata granting module (53).

[0092] The access authority verification module (51) can perform the function of determining whether access to a specific persona attribute or basis is possible by comprehensively considering the identity of the requesting subject, the relationship with the representative user, the public level and the scope of permission to use.

[0093] For example, the scope of accessible data can be set differentially depending on the type of requesting entity, such as family, friends, or external services, and additional restrictions can be applied based on the purpose or risk level of the activity.

[0094] The data conversion module (52) can perform various conversion operations on persona attributes or grounds that are restricted from access.

[0095] For example, for sensitive information, de-identification processing can be performed by masking some values ​​or removing identifiable elements, and detailed information can be categorized or summarized and provided in a brief form.

[0096] In addition, for attributes that are prohibited from being disclosed, they can be handled by completely removing them.

[0097] The authorization metadata granting module (53) can perform the function of granting the scope of authorization applied to the transformed persona evidence, the availability conditions, and the transformation history information in the form of metadata.

[0098] This allows tracking under what authorization conditions the data was generated during subsequent interactions or proxy actions.

[0099] In addition, the authority conversion unit (50) can generate different forms of conversion results depending on the requesting subject and the context of the action, even for the same persona evidence graph, and can be configured to provide customized information that reflects multiple access levels for a single user.

[0100] For example, the source text and detailed attribute information may be provided to the user as is, but may be configured to be provided to external users only in a summarized or de-identified form.

[0101] Accordingly, the authority conversion unit (50) performs a core function that simultaneously satisfies personal information protection and authority-based data utilization by performing precise conversion according to access rights on the information included in the persona evidence graph.

[0102] Figure 3 is a block diagram showing the authority transformation flow in which persona evidence is transformed according to the authority scope of the requesting subject.

[0103] Referring to FIG. 3, when identification information and relationship information of the requesting subject are input, the access authority verification module (51) determines whether access to persona attributes or grounds is possible, and if access is restricted, the data conversion module (52) performs conversions such as masking, de-identification, categorization, or summarization, and then the authority metadata granting module (53) generates persona evidence converted with applied authority conditions and conversion history.

[0104] Next, the agent configuration part (60) will be described.

[0105] The agent configuration unit (60) is a configuration that creates or updates a user-attributed agent based on the persona evidence graph or the converted persona evidence.

[0106] In addition, the agent configuration unit (60) may be configured such that the user-attributed agent includes at least one of a basic generation model, a user-attributed system prompt, a search-based basis reference, a behavior policy module, and a risk assessment module, and the user-attributed system prompt may be configured such that it includes at least one of a priority of persona attributes, speech characteristics, a non-disclosure rule, or an approval requirement rule.

[0107] At this time, the agent configuration unit (60) may be configured to include an agent profile generation module (61), a system prompt generation module (62), and a policy / risk module linkage module (63).

[0108] The agent profile generation module (61) can perform the function of generating an agent profile that reflects the unique characteristics of the user from the persona evidence graph or the converted persona evidence.

[0109] For example, an internal state expression representing the user can be constructed by combining interests, decision-making criteria, tone of voice, and writing style characteristics.

[0110] The system prompt generation module (62) can perform the function of generating a user-associated system prompt based on the agent profile.

[0111] The above system prompts may include the priority of persona attributes, tone and politeness during responses, targets for non-disclosure, and constraints on actions requiring approval, and can guide the generative model to maintain a consistent persona.

[0112] The policy / risk module linkage module (63) can perform the function of linking the behavior policy module and the risk assessment module to the user-associated agent.

[0113] This allows the agent to be configured to operate by reflecting policies and risk criteria by behavior type, not only when generating responses but also when performing proxy actions.

[0114] In addition, the agent configuration unit (60) can update the user-attributed agent when new persona attributes or grounds are added or user feedback is reflected, and can be configured to maintain consistency with the existing agent state when updating.

[0115] For example, if a specific user's preferences change or new interests are added, the system prompt or agent profile can be updated to reflect this in subsequent interactions.

[0116] Accordingly, the agent configuration unit (60) creates an agent that reflects user characteristics based on the persona evidence graph and continuously updates it, thereby forming a core configuration that enables consistent and personalized actions to be performed during subsequent interactions and proxy actions.

[0117] Next, the interaction control unit (70) will be described.

[0118] The interaction control unit (70) is configured to control the user-associated agent to operate in at least one interaction mode among a single proxy mode, a two-way conversation mode, a representative discussion mode, or a third-party persona conversation mode.

[0119] In addition, the interaction control unit (70) may be configured to select discussion participation agents based on at least one of user attribute diversity, issue suitability, scope of disclosure, or user-specified priority when multiple user-affiliated agents participate in a discussion in the representative discussion mode.

[0120] In addition, the interaction control unit (70) may be configured to verify a pre-registered permission package for a third party in the third-party persona conversation mode and to activate the third-party persona conversation mode only if at least one of the permission period, purpose of use, type of allowed data, and prohibited act included in the permission package is valid.

[0121] At this time, the interaction control unit (70) may be configured to include a mode selection module (71), a discussion participation selection module (72), and a speech / exchange control module (73).

[0122] The mode selection module (71) can perform the function of selecting an appropriate interaction mode according to user requests, the purpose of interaction, and the current situation.

[0123] For example, if a simple response is required, you can select the single proxy mode, and if an exchange of opinions between users is required, you can select the two-way conversation mode or the representative discussion mode.

[0124] The discussion participation selection module (72) can perform the function of selecting a user-associated agent to participate in the representative discussion mode.

[0125] At this time, an appropriate agent can be selected by considering the diversity of user attributes, suitability with the issue, scope of disclosure, and custom priority.

[0126] The speech / exchange control module (73) can perform the function of controlling the speech order between agents, the message exchange method, and the response flow according to the selected interaction mode.

[0127] For example, in discussion mode, the order of speech can be controlled in a cyclical manner, or the message flow can be adjusted to allow for focused discussion on specific issues.

[0128] In addition, the interaction control unit (70) can restrict the range of information that each agent can access based on the converted persona evidence transmitted from the authority conversion unit (50), thereby controlling the interaction process so that information exposure outside the authority does not occur.

[0129] For example, a specific agent may be restricted to speaking based only on summarized information, while another agent may be configured to utilize more detailed grounds.

[0130] Accordingly, the interaction control unit (70) selects various interaction modes according to the situation and controls the flow of participation and speech between agents, thereby performing the role of ensuring effective interaction while complying with the scope of authority.

[0131] Figure 5 is a block diagram illustrating the process of performing a discussion and deriving results in a representative discussion mode in which multiple user-associated agents participate.

[0132] Referring to FIG. 5, when a discussion topic or question is entered, a plurality of user-associated agents are selected by the discussion participation selection module (72), the speaking order and message exchange of each agent are controlled by the speaking / exchange control module (73), and the speaking results are accumulated and compared to derive the final discussion result.

[0133] Figure 6 is a block diagram showing the flow of verifying permission packages and determining whether the conversation mode is active in third-party persona conversation mode.

[0134] Referring to Fig. 6, when a third-party persona conversation request is input, a pre-registered permission package is verified, and the third-party persona conversation mode is activated only when the permission period, purpose of use, type of allowed data, and conditions for prohibited acts are satisfied, and the mode is blocked when the above conditions are not satisfied.

[0135] Next, the proxy action control unit (80) will be described.

[0136] The proxy action control unit (80) is a configuration that controls the user-associated agent to perform at least one proxy action, such as creating a message, proposing a schedule, creating an approval request, calling an external API, or executing a workflow.

[0137] In addition, the proxy behavior control unit (80) may be configured to classify the proxy behavior into one of the following based on a policy table for each type of behavior: an action that can be performed automatically, an action that requires user approval, or a blocking action; generate an approval request if the risk level of the action is greater than or equal to a threshold value; and control the proxy behavior to be performed automatically if the risk level of the action is less than the threshold value and automatic execution is permitted by the policy table for each type of behavior.

[0138] In addition, the proxy behavior control unit (80) may be configured to calculate the risk of the said behavior based on at least one of data sensitivity, external system influence, monetary effect, legal effect, user absence status, or lack of reliability of grounds.

[0139] At this time, the proxy behavior control unit (80) may be configured to include a behavior classification module (81), a risk assessment module (82), an approval request generation module (83), and an execution control module (84).

[0140] The behavior classification module (81) can identify the type of proxy action requested for execution and classify the action into an action that can be automatically executed, an action requiring user approval, or a blocking action according to a predefined policy table for each action type.

[0141] For example, simple message generation can be classified as an action that can be performed automatically, whereas a payment request involving money can be classified as an action requiring user approval.

[0142] The risk assessment module (82) can perform the function of calculating the risk associated with the above-mentioned proxy action.

[0143] At this time, a risk score can be calculated by comprehensively considering data sensitivity, external system impact, financial effects, legal effects, user absence, and the lack of reliability of the basis.

[0144] For example, the act of transmitting data containing sensitive personal information can be assessed as high risk.

[0145] The approval request generation module (83) can perform the function of generating an approval request to a user when the above risk assessment result is greater than or equal to a threshold value or is classified as requiring approval by the policy table for each type of behavior.

[0146] The above approval request may include the details of the action to be performed, the basis used, expected results, and risk information.

[0147] The execution control module (84) can perform the function of controlling the actual proxy action through an external API call or workflow execution when the above action is classified as an action that can be automatically performed or when execution is allowed after user approval.

[0148] In addition, the proxy action control unit (80) can be configured to limit the range of information available for use when performing proxy actions based on the persona evidence converted by the authority conversion unit (50), thereby preventing the use of data outside the scope of authority.

[0149] For example, data transmitted to external services may be restricted to a de-identified or summarized form, and may be configured to be transmitted with sensitive information automatically removed.

[0150] Accordingly, the proxy action control unit (80) determines whether to perform proxy actions based on policies and risk assessments for each type of action, and controls them to be executed safely through an approval process when necessary, thereby enabling reliable and controllable proxy actions even in the absence of a user.

[0151] Figure 4 is a block diagram showing the control flow for a proxy action that assesses the risk and determines whether to approve when a proxy action is performed by a user-attributed agent.

[0152] Next, the moderation unit (90) is described.

[0153] The moderation unit (90) is configured to evaluate the result of the interaction mode or the result of the proxy action to generate a final response or final action result.

[0154] In addition, the moderation unit (90) may be configured to generate the final response or the final action result based on at least one of persona match, grounds reliability, authority suitability, behavioral risk, issue coverage, or user constraint compliance.

[0155] At this time, the moderation unit (90) may be configured to include an evaluation indicator calculation module (91) and a result generation module (92).

[0156] The evaluation indicator calculation module (91) can perform the function of calculating various evaluation indicators for candidate results generated during the above interaction or proxy action process.

[0157] For example, quantitative or qualitative indicators can be used to calculate how well the generated response matches the user persona, whether the reliability of the evidence used is sufficient, and whether the scope of authority is adhered to.

[0158] The result generation module (92) can perform the function of selecting the optimal result among multiple candidate results based on the above evaluation indicators, or combining multiple results if necessary to generate a final response or final action result.

[0159] In addition, explainability can be enhanced by generating a summary of the basis or explanatory information for the selected results.

[0160] In addition, the moderation unit (90) can be configured to re-verify the authority conditions defined in the authority conversion unit (50) to further block the possibility of authority violation at the final output stage.

[0161] Therefore, the moderation department (90) determines the final result based on various evaluation criteria, thereby performing the role of simultaneously ensuring response quality and suitability of the behavior.

[0162] Next, the audit record book (100) is described.

[0163] The audit logbook (100) is configured to store persona attributes, grounds, transformation results, approval history, and execution logs used in the process of performing the interaction mode or the proxy action.

[0164] At this time, the audit record book (100) may be configured to include a log collection module (101), a basis / conversion history storage module (102), and an approval history storage module (103).

[0165] The log collection module (101) can perform the function of collecting various logs including interaction messages, system events, results of proxy action execution, and visual information.

[0166] The basis / conversion history storage module (102) can perform the function of storing the persona attribute used in a specific response or action, the basis of the attribute, and the conversion history applied by the authority conversion unit (50) in conjunction.

[0167] The approval history storage module (103) can perform the function of storing an approval history including the time of creation of a user approval request, whether it is approved or rejected, approval conditions, and execution results after approval.

[0168] In addition, the audit record book (100) can be configured to maintain the relationship between each log item so that it is possible to trace back what basis and what transformation process was used for a specific result.

[0169] Therefore, the audit record book (100) plays a role in ensuring post-verification and auditing possibilities by systematically storing the history of the entire process.

[0170] Figure 7 is a conceptual diagram showing the log structure and trace information stored in the audit log.

[0171] Next, the learning update unit (110) will be described.

[0172] The learning update unit (110) is configured to update the persona evidence graph or the policy table by behavior type by reflecting user feedback, approval results, cancellation results, modification instructions, or satisfaction data after interaction.

[0173] At this time, the learning update unit (110) may be configured to include a feedback collection module (111) and a profile / policy update module (112).

[0174] The feedback collection module (111) can perform the function of collecting evaluations, modification requests, approval / rejection history, and satisfaction information entered by the user.

[0175] The profile / policy update module (112) can perform the function of adjusting the reliability or priority of persona attributes based on the collected data, or updating the reference values ​​and classification rules of the policy table by behavior type.

[0176] In addition, the learning update unit (110) can be configured to automatically optimize the policy by analyzing repetitive feedback patterns or to dynamically adjust the approval criteria under specific conditions.

[0177] Therefore, the learning update unit (110) continuously reflects the operation results of the system to improve personas and policies, thereby playing a role in improving personalization accuracy and control consistency in the long term.

[0178] Next, the version control department (120) will be described.

[0179] The version control unit (120) is configured to generate and save a snapshot at every time the persona evidence graph is changed, and to determine the valid attribute when a conflict occurs between the persona attribute candidates.

[0180] At this time, the version control unit (120) may be configured to include a snapshot creation module (121) and a conflict resolution module (122).

[0181] The snapshot generation module (121) can perform the function of saving the state of the persona evidence graph in time units so that the persona state at a specific point in time can be restored.

[0182] The conflict resolution module (122) can perform the function of determining valid attributes based on at least one of recency, source reliability, whether user-specified modifications or repeated observation frequency when conflicting persona attribute candidates exist.

[0183] In addition, the version control unit (120) may be configured to analyze the history of persona changes by comparing past snapshots and the current state, or to support rollback to a specific point in time.

[0184] Therefore, the version control unit (120) performs the role of maintaining the consistency and reliability of persona information by performing history management and conflict resolution of the persona evidence graph.

[0185] In the preceding section, we described the system for proxy behavior and authority control interactions of a user-attributed agent using a multimodal evidence-based persona graph. Below, we will describe the method for proxy behavior and authority control interactions of a user-attributed agent using a multimodal evidence-based persona graph.

[0186] FIG. 8 is a flowchart showing the processing flow of a method for surrogate behavior and authority control interaction of a user-attributed agent using a multimodal basis-based persona graph according to an embodiment of the present invention.

[0187] Referring to FIG. 8, a method for surrogate behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph according to an embodiment of the present invention includes the steps of: collecting multimodal user data (S10); preprocessing the multimodal user data (S20); extracting persona attribute candidates and attribute evidence metadata (S30); generating a persona evidence graph (S40); converting persona evidence according to authority (S50); generating or updating a user-attributed agent (S60); controlling an interaction mode (S70); controlling to perform surrogate behavior (S80); and recording the result of the performance (S90).

[0188] First, the step of collecting multimodal user data (S10) is a step of collecting multimodal user data including documents, images, conversation history, user input, or behavior records associated with the user.

[0189] Next, the step of preprocessing the multimodal user data (S20) is a step of preprocessing the multimodal user data into a form suitable for attribute extraction.

[0190] Next, the step of extracting persona attribute candidates and attribute basis metadata (S30) is a step of extracting persona attribute candidates and attribute basis metadata corresponding to said persona attribute candidates from said preprocessed multimodal user data.

[0191] Next, the step of generating a persona evidence graph (S40) is to generate a persona evidence graph by associating the persona attribute candidates with a user persistent identifier, source, trust level, visual information, public rating, and usage permission scope.

[0192] Next, the step of converting persona evidence according to authority (S50) is a step of generating converted persona evidence by converting the persona attributes or grounds included in the persona evidence graph according to the relationship between the requesting subject and the user identified by the user persistent identifier, the disclosure level, the usage permission scope, and the behavioral risk level.

[0193] Next, the step of creating or updating the user attribution agent (S60) is the step of creating or updating the user attribution agent based on the persona evidence graph or the transformed persona evidence.

[0194] Next, the step of controlling the interaction mode (S70) is to control the user-attributed agent to operate in at least one interaction mode among a single proxy mode, a two-way conversation mode, a representative discussion mode, or a third-party persona conversation mode using the converted persona evidence.

[0195] Next, the step of controlling to perform a proxy action (S80) is a step of controlling the user-associated agent to perform at least one proxy action among message creation, schedule proposal, approval request creation, external API call, or workflow execution.

[0196] Finally, the step of recording the execution result (S90) is a step of storing the persona attributes, grounds, transformation results, approval history, and execution logs used in the process of performing the interaction mode or the proxy action.

[0197] The specific configuration and operation of each of the above steps have been explained in detail in the description of the system configuration mentioned above, so a redundant explanation thereof will be omitted.

[0198] In addition, the method for surrogate behavior and authority control interaction of a user-attributed agent using a multimodal basis-based persona graph according to one embodiment of the present invention may be implemented in the form of program code executable by a computing device, and such program code may be stored and provided on a non-transient computer-readable recording medium.

[0199] The system, method, and recording medium for performing the interaction of proxy behavior and authority control of a user-attributed agent using a multimodal basis-based persona graph described above and illustrated in the drawings are merely one embodiment for carrying out the present invention and should not be interpreted as limiting the technical scope of the present invention. The scope of protection of the present invention is determined solely by the matters described in the following claims, and embodiments that are improved and modified without departing from the gist of the present invention shall be deemed to fall within the scope of protection of the present invention insofar as they are obvious to those skilled in the art to which the present invention belongs. Explanation of the symbols

[0200] 10: Input Collection Unit 20: Preprocessing section 30: Attribute Extraction Unit 40: Persona Graph Generation Section 50: Authority conversion unit 60: Agent Configuration 70: Interaction control unit 80: Proxy Behavior Control Unit 90: Moderation Department 100: Audit Record Book 110: Learning Update Department 120: Version Control Department S10: Step for collecting multimodal user data S20: Step of preprocessing the above multimodal user data S30: Step to extract persona attribute candidates and attribute basis metadata S40: Step to generate the Persona Evidence Graph S50: Step to transform persona evidence according to authority S60: Step to create or update user attribution agents S70: Step for controlling interaction mode S80: Step for controlling the performance of proxy actions S90: Step for recording execution results

Claims

Claim 1 An input collection unit that collects multimodal user data including documents, images, conversation history, user input, or behavior records associated with a user; a preprocessing unit that preprocesses the multimodal user data into a form suitable for attribute extraction; an attribute extraction unit that extracts persona attribute candidates and attribute basis metadata corresponding to the persona attribute candidates from the preprocessed multimodal user data; a persona graph generation unit that generates a persona evidence graph by linking the persona attribute candidates with a user persistent identifier, source, trust level, visual information, disclosure level, and usage permission scope; an authorization conversion unit that generates converted persona evidence by converting the persona attributes or basis included in the persona evidence graph according to the relationship between the requesting subject and the user identified by the user persistent identifier, the disclosure level, the usage permission scope, and behavioral risk level; an agent configuration unit that generates or updates a user-attributed agent based on the persona evidence graph or the converted persona evidence; and the user-attributed agent is in a sole proxy mode, two-way conversation mode, representative discussion mode, or third-party persona An interaction system for proxy actions and authorization control of a user-attributed agent using a multimodal basis-based persona graph, characterized by comprising: an interaction control unit that controls the agent to operate in at least one interaction mode among conversation modes; a proxy action control unit that controls the agent to perform at least one proxy action among message generation, schedule proposal, approval request generation, external API call, or workflow execution; and an audit log unit that stores persona attributes, basis, transformation results, approval history, and execution logs used in the interaction mode or the process of performing the proxy action. Claim 2 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal basis-based persona graph, characterized in that, in claim 1, the input collection unit is configured such that the collected multimodal user data includes at least one of a text document, an image or a document containing an image, a messenger conversation, an email, a voice conversation transcript, a self-introduction input, a profile question response, a schedule information, a work history, a purchase history, an approval history, or external profile data. Claim 3 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal basis-based persona graph, characterized in that, in claim 1, the preprocessing unit is configured to perform at least one of document optical character recognition (OCR), sentence separation, tokenization, named entity recognition, conversational speaker separation, image feature extraction, personal identification information detection, sensitive information tagging, time-series sorting, or duplicate removal on the multimodal user data. Claim 4 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph, characterized in that, in claim 1, the attribute extraction unit extracts at least one of demographic attributes, occupation, professional field, interests, hobbies, value judgment criteria, preferences, dislikes, decision priority, speech characteristics, writing style characteristics, length characteristics, politeness characteristics, response patterns by relationship group, or repetitive behavior rules from the preprocessed multimodal user data as a persona attribute candidate, and generates attribute basis metadata for each persona attribute candidate, including an attribute name, attribute value, source of basis, reliability score, extraction time, and extraction method identifier. Claim 5 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph, wherein, in claim 1, the persona graph generation unit comprises a user node, an attribute node, a basis node, and a relationship edge, wherein the basis node comprises a source identifier, a time of creation, a reliability, a disclosure level, and a usage permission scope, and wherein the relationship edge is configured to represent at least one of a relationship where the basis supports the attribute, a relationship where the basis conflicts with the attribute, a relationship where the attribute is derived from a parent attribute, a relationship that can be disclosed to a specific group of relationships, or a relationship where a specific usage permission period expires. Claim 6 A proxy behavior and authority control interaction system of a user-attributed agent using a multimodal basis-based persona graph, characterized in that, in claim 1, the authority conversion unit is configured to generate the converted persona evidence by performing at least one of masking, de-identification, categorization, summarization, or removal of prohibited-disclosure attributes on the basis or attribute when the requesting subject cannot directly access the basis or attribute included in the persona evidence graph. Claim 7 A proxy behavior and authority control interaction system of a user attribution agent using a multimodal evidence-based persona graph, characterized in that, in claim 1, the agent component generates or updates the user attribution agent including at least one of a basic generative model, a user attribution system prompt, a search-based evidence referencer, a behavior policy module, and a risk assessment module based on the persona evidence graph or the transformed persona evidence, and the user attribution system prompt is configured to include at least one of a priority of persona attributes, speech characteristics, a non-disclosure rule, or an approval requirement rule. Claim 8 A system for proxy behavior and authority control interaction of user-associated agents using a multimodal basis-based persona graph, characterized in that, in claim 1, the interaction control unit is configured to select discussion participation agents based on at least one of user attribute diversity, issue suitability, scope of disclosure, or user-specified priority when multiple user-associated agents participate in a discussion in the representative discussion mode. Claim 9 A system for the interaction of proxy behavior and authorization control of a user-attributed agent using a multimodal basis-based persona graph, characterized in that, in claim 1, the proxy behavior control unit classifies the proxy behavior into one of an action that can be performed automatically, an action requiring user approval, or a blocking action based on a policy table by action type, generates an approval request when the risk level of the action is greater than or equal to a threshold value, and controls the proxy behavior to be performed automatically when the risk level of the action is less than the threshold value and automatic execution is permitted by the policy table by action type. Claim 10 In claim 9, the proxy behavior control unit is configured such that the behavior risk is calculated based on at least one of data sensitivity, external system influence, monetary effect, legal effect, user absence status, or lack of reliability of grounds, in a proxy behavior and authority control interaction system of a user-attributed agent using a multimodal basis-based persona graph. Claim 11 A proxy behavior and authority control interaction system of a user-attributed agent using a multimodal basis-based persona graph, characterized in that, in claim 1, the interaction control unit is configured to verify a pre-registered permission package for a third party in the third-party persona conversation mode and to activate the third-party persona conversation mode only when at least one of the permission period, purpose of use, type of allowed data, and prohibited act included in the permission package is valid. Claim 12 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph, wherein, in claim 1, the system further comprises a moderation unit that evaluates the result of the interaction mode or the result of the proxy behavior to generate a final response or a final action result; and wherein the moderation unit is configured to generate the final response or the final action result based on at least one of persona consistency, evidence reliability, authority suitability, behavioral risk, issue coverage, or user constraint compliance. Claim 13 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph, characterized in that, in claim 9, the system is further configured to include a learning update unit that updates the persona evidence graph or the policy table by behavior type by reflecting user feedback, approval results, cancellation results, modification instructions, or satisfaction data after interaction. Claim 14 A system for proxy behavior and authority control interaction of a user-attributed agent using a multimodal evidence-based persona graph, wherein, in claim 1, the system further comprises a version control unit that generates and stores a snapshot at each point of change of the persona evidence graph; and wherein the version control unit is configured to determine valid attributes based on at least one of recency, source reliability, whether user-specified modifications have been made, or frequency of repeated observations when a conflict occurs between the persona attribute candidates. Claim 15 A method for surrogate behavior and authorization control interactions of a user-attributed agent using a multimodal evidence-based persona graph performed by a computing device, comprising: a step of collecting multimodal user data including documents, images, conversation history, user input, or behavior records associated with a user; a step of preprocessing the multimodal user data into a form suitable for attribute extraction; a step of extracting persona attribute candidates and attribute evidence metadata corresponding to the persona attribute candidates from the preprocessed multimodal user data; a step of generating a persona evidence graph by linking the persona attribute candidates with a user persistent identifier, source, trust level, visual information, disclosure level, and usage scope; a step of generating transformed persona evidence by transforming the persona attributes or evidence included in the persona evidence graph according to the relationship between the requesting subject and the user identified by the user persistent identifier, the disclosure level, the usage scope, and behavioral risk level; a step of generating or updating a user-attributed agent based on the persona evidence graph or the transformed persona evidence; and a step in which the user-attributed agent, using the transformed persona evidence, performs a sole A method for controlling proxy actions and authorization control interactions of a user-attributed agent using a multimodal basis-based persona graph, characterized by comprising: a step of controlling the agent to operate in at least one interaction mode among proxy mode, two-way conversation mode, representative discussion mode, or third-party persona conversation mode; a step of controlling the user-attributed agent to perform at least one proxy action among message creation, schedule proposal, approval request creation, external API call, or workflow execution; and a step of storing persona attributes, basis, transformation results, approval history, and execution logs used in the interaction mode or the process of performing the proxy action. Claim 16 A non-transient computer-readable recording medium on which instructions for performing the method described in paragraph 15 are recorded.

Citation Information

Patent Citations

  • Persona-Security Graph-based dynamic security management system and method thereof

    KR102910983B1

  • Implementing personal knowledge graphs

    US20250007723A1

  • Method for providing services for one or more persons

    US20250173658A1