Method and apparatus for providing artificial intelligence-based conference service

Abstract

According to an embodiment, provided is a computer-implemented method performed by a conference service apparatus comprising a memory, a processor, and one or more programs stored in the memory and configured to be executed by the processor, the computer-implemented method comprising the steps of: acquiring, by an agent, conversations received from conference participants who participate in an online conference room including the agent connected to a language model; selecting, by the agent, a target conversation from among the conversations, the target conversation including a conversation requiring a response using local internal data; providing, by the agent, the target conversation as input data to the language model, the language model generating, for a natural language input, a natural language response by using the local internal data; acquiring, by the agent, output data for the input data from the language model; and providing, by the agent on the basis of the output data, a conversation response to the target conversation to participant terminals of the conference participants.

Description

Method and device for providing artificial intelligence-based conference services

[0001] The present disclosure relates to a method and device for providing an artificial intelligence-based conference service.

[0002] With the advancement of the Internet, various technologies have been developed for users to conduct electronic meetings. Among the most widely used conferencing technologies, messenger technologies are widely known for enabling remote connection of users' communication environments and real-time information exchange, enabling meetings between users. Furthermore, video conferencing technologies based on video messengers, which combine messenger and vision technologies, are being developed. These technologies enable real-time video conferencing, allowing users in remote locations to share chat, voice, and video content.

[0003] With the recent advancement of AI technology, various methods are being attempted to integrate AI into conferencing technology. For example, there are technologies that support the meeting environment by allowing AI-based virtual assistants to record meetings or by adjusting environmental factors of video conferences according to user commands.

[0004] However, although these conventional technologies have advantages in terms of improving user convenience by supporting a smooth meeting environment or replacing replaceable user actions, they have limitations in that they do not contribute much to the effectiveness of the meeting in achieving the meeting's purpose.

[0005] Accordingly, the need to develop artificial intelligence-based conferencing technology to enhance meeting effectiveness is growing.

[0006] The problem to be solved according to one embodiment includes resolving the above-described problems and / or limitations.

[0007] Another challenge to be addressed in one embodiment involves overcoming limitations in response reliability, such as hallucination, which frequently occurs in generative AI.

[0008] Another problem to be solved, according to another embodiment, includes providing a conference service that can enhance the effectiveness of a conference by supporting the achievement of the conference's goals or the resolution of problems in specialized fields based on artificial intelligence.

[0009] However, the problems to be solved by the present invention are not limited to those mentioned above, and other problems to be solved that are not mentioned can be clearly understood by a person having ordinary skill in the art to which the present invention pertains from the description below.

[0010] A computer-implemented method performed by a conference service device including a memory, a processor, and one or more programs stored in the memory and configured to be executed by the processor according to one embodiment of the present disclosure, the computer-implemented method may include: a step in which an agent acquires conversations received from conference participants participating in an online conference room including an agent connected to a language model; a step in which the agent selects a target conversation from among the conversations; the target conversation includes a conversation requiring a response using local internal data; a step in which the agent provides the target conversation as input data to the language model; the language model generates a natural language response using the local internal data for a natural language input; a step in which the agent acquires output data for the input data from the language model; and a step in which the agent provides a conversation response for the target conversation to a participant terminal of the conference participant based on the output data.

[0011] In one embodiment, the agent is a conference participant object that participates in the online conference room together with the conference participants, and may be granted the same conversation input and conversation reception authority as the conference participants.

[0012] In one embodiment, the method further includes a step of creating the online conference room based on a conference opening request from a host terminal, and in the step of creating the online conference room, it is possible to determine whether the agent will participate.

[0013] In one embodiment, the computer-implemented method may further include, after the participation of the agent is determined, the step of the conference service device obtaining conference information of the online conference room and obtaining external data associated with the conference information before the online conference room is started; and the step of performing additional learning on the language model using the external data associated with the conference information.

[0014] In one embodiment, the step of selecting the target conversation is characterized in that the agent selects the target conversation using a preset selection rule, and the selection rule may be based on at least one selected from the group consisting of a type of conversation, an intention of the conversation, whether a counterpart conference participant is nominated, whether an AI (Artificial Intelligence) conference participant is nominated, a context of conversations, a time interval of conversations, a number of conversations during a predetermined time period, and whether text associated with the local internal data is included.

[0015] In one embodiment, in the step of providing the target conversation to the language model, the agent provides the target conversation and the conversations to the language model, and in the step of the agent obtaining output data from the language model, the language model can generate the output data based on additional learning using the conversations.

[0016] In one embodiment, the language model is trained to select a conversation requiring a response based on a plurality of conversations and the necessity of the response, and in the step of selecting the target conversation, the agent can transmit the received conversations to the language model.

[0017] In one embodiment, the local internal data is assigned a security level associated with the user's access rights, and the access rights may be set by considering at least one selected from the group consisting of rank, position, document security level, department, employment type, business location, and affiliate.

[0018] In one embodiment, the language model may generate a natural language response to a natural language input by using local internal data of a security level corresponding to the access rights granted to each meeting participant among the local internal data.

[0019] In one embodiment, the local internal data includes local internal data associated with an organization to which at least some of the meeting participants belong, and the local internal data associated with the organization can be categorized based on one or more types selected from the group consisting of human resources, finance, accounting, legal, management, research, development, production, and sales.

[0020] In one embodiment, the language model may be a generative artificial intelligence (AI) model.

[0021] In one embodiment, the generative AI model may be a pre-trained model using the local internal data.

[0022] In one embodiment, the generative AI model may be further trained using meeting information and / or conversations from the online conference room.

[0023] In one embodiment, the local internal data may include conversations from meetings prior to the creation of the online conference room.

[0024] In one embodiment, the generative AI model may be further trained using conversations from meetings prior to the creation of the online conference room.

[0025] In one embodiment, the meeting information of the online conference room may include at least one selected from the group consisting of (i) a meeting period, (ii) meeting content, (iii) information on meeting participants, (iv) meeting classification information, and (v) a meeting history indicating a historical relationship between the online conference room and one or more other online conference rooms associated with the online conference room.

[0026] In one embodiment, the generative AI model can be further trained using training data containing predefined reference conversations.

[0027] In one embodiment, the generative AI model may be a model learned using specialized field data associated with a preset specialized field.

[0028] In one embodiment, the step of the agent obtaining the output data may include: the agent obtaining a search result for local internal data associated with the target conversation from the local internal data; the agent generating a prompt for instructing the generative AI model to generate the output data for the target conversation using the search result; and the agent providing the prompt to the generative AI model, thereby obtaining the output data from the generative AI model.

[0029] In one embodiment, the prompt may be to instruct the generative AI model to generate the output data by additionally utilizing the conversations and / or meeting information of the online conference room.

[0030] In one embodiment, the step of selecting the target conversation may include the agent selecting one of a first target conversation requiring a response using local internal data and a second target conversation requiring a response using specialized field data from among the conversations, and the language model may include a first model that generates a natural language response using the local internal data for a natural language input and a second model that is trained to generate a natural language response using specialized field data associated with a preset specialized field for a natural language input, and in the step of providing the target conversation as input data, the agent may provide the first target conversation as the input data to the first model when the first target conversation is selected, and may provide the second target conversation as the input data to the second model when the second target conversation is selected.

[0031] In one embodiment, the language model is stored in the memory or in an external device connected to the conference service device via a network, and can be executed by the agent.

[0032] In one embodiment, the language model may generate the output data in at least one response form selected from the group consisting of questions, answers, opinions, and summaries.

[0033] In one embodiment, the response form may be determined by 1) the language model based on the target conversation, or 2) the agent based on the target conversation.

[0034] In one embodiment, the step of providing the conversation response may provide the conversation response to the participant terminal in the form of a message by an AI (Artificial Intelligence) conference participant that is visually distinct from the conference participants.

[0035] In one embodiment, the step of providing the conversation response may determine whether to provide the conversation response based on the access rights granted to each of the meeting participants.

[0036] In one embodiment, the step of providing the conversation response may provide the conversation response obtained by using local internal data to which a security level corresponding to the access rights of each conference participant is assigned to each participant terminal of the conference participants.

[0037] In one embodiment, the step of providing the dialogue response may include providing, together with the dialogue response, the basis data used to obtain the output data from among the local internal data and / or an access link to the basis data.

[0038] In one embodiment, the step of providing the conversation response may provide the conversation response only to the participant terminal when the target conversation is a conversation for which a response has been individually requested by the participant terminal.

[0039] In one embodiment, the computer-implemented method may further include a step in which the agent obtains a meeting result report including summary results for each of a plurality of preset criteria items from the conversations; the plurality of criteria items include at least one selected from the group consisting of a meeting summary item, a meeting content item, a main conversation collection item, a task item, and a next meeting item; and a step in which the agent provides the meeting result report to the participant terminal.

[0040] In one embodiment, the agent may further include: 1) scheduling the next meeting item based on an approval response from the meeting participant for the next meeting item; and / or 2) tracking management of the work item based on information about the work progress and / or work result received from a participant terminal associated with the work item.

[0041] In one embodiment, the computer-implemented method may further include, before receiving the conversations, a step of receiving a conference opening request including a conference period, conference content, and information on conference participants from a host terminal; a step of creating an online conference room including a conversation thread based on conversation input of the conference participants based on the conference opening request; and a step of controlling the conference to proceed with non-real-time participation of the conference participants by providing the online conference room through a network in response to a request from each participant terminal of the conference participants during the conference period.

[0042] In one embodiment, the method further comprises obtaining video data related to the meeting content, and the online conference room may further include the video data.

[0043] A conference service device according to one embodiment of the present disclosure comprises: a memory in which a language model for generating a natural language response using local internal data for a natural language input and at least one command are stored; and a processor configured to execute the at least one command stored in the memory, wherein the at least one command, when executed by the processor, causes the processor to perform the following method, wherein the method comprises: obtaining conversations received from conference participants participating in an online conference room including an agent connected to the language model; selecting a target conversation among the conversations; the target conversation including a conversation requiring a response using local internal data; providing the target conversation as input data to the language model; obtaining output data for the input data from the language model; and providing a conversation response for the target conversation to a participant terminal of the conference participant based on the output data.

[0044] In one embodiment of the present disclosure, a computer program stored in a computer-readable non-transitory recording medium includes instructions that, when executed by one or more processors, cause the one or more processors to perform a computer-implemented method, the computer-implemented method may include: a step in which the agent acquires conversations received from conference participants participating in an online conference room including an agent connected to a language model; a step in which the agent selects a target conversation from among the conversations; the target conversation includes a conversation requiring a response using local internal data; a step in which the agent provides the target conversation as input data to the language model; the language model generates a natural language response using the local internal data for a natural language input; a step in which the agent obtains output data for the input data from the language model; and a step in which the agent provides a conversation response for the target conversation to a participant terminal of the conference participant based on the output data.

[0045] In one embodiment of the present disclosure, a computer-readable non-transitory recording medium storing a computer program includes instructions that, when executed by one or more processors, cause the one or more processors to perform a computer-implemented method, the computer-implemented method may include: a step in which the agent acquires conversations received from conference participants participating in an online conference room including an agent connected to a language model; a step in which the agent selects a target conversation from among the conversations; the target conversation includes a conversation requiring a response using local internal data; a step in which the agent provides the target conversation as input data to the language model; the language model generates a natural language response using the local internal data for a natural language input; a step in which the agent obtains output data for the input data from the language model; and a step in which the agent provides a conversation response for the target conversation to a participant terminal of the conference participant based on the output data.

[0046] According to one embodiment of the present disclosure, a conference service is provided in which, in an online conference environment where multiple people are conversing, AI-based colleagues with the ability to participate in discussions on meeting agendas and / or AI-based experts with the ability to consult for problem-solving in their specialized fields participate in the meeting. Accordingly, the meeting's objectives can be more easily achieved.

[0047] According to one embodiment of the present disclosure, an AI conference participant can be provided that provides reliable responses using local internal data during a meeting. Furthermore, by selectively responding to conversations requiring a response, the AI ​​conference participant can participate naturally in multi-party conference conversations without disrupting the flow of conversation among conference participants. Furthermore, the AI ​​conference participant can be provided as a colleague AI that is knowledgeable about local internal data and capable of engaging in conversation, enabling equal participation in the meeting with other conference participants. Accordingly, conference participants can utilize necessary information in a timely manner based on conversations with their AI colleague regarding local internal data during the meeting, thereby facilitating more effective meetings.

[0048] According to another embodiment of the present disclosure, an AI conference participant can be provided that provides specialized responses using specialized data related to a given field of expertise. Based on intensive learning of specialized knowledge and expertise specific to a specific field of expertise, the AI ​​conference participant can be provided as an expert AI capable of providing specialized consulting for various situations requiring problem solving related to that field of expertise. Accordingly, when a conversation with a specialist in a specific field is required during the meeting, the meeting participants can invite the expert AI to provide timely problem resolution, thereby facilitating more effective meetings.

[0049] According to another embodiment of the present disclosure, an AI conference participant can be provided that autonomously determines which conversations require a response and selectively participates in them without disrupting the flow of conversation among meeting participants. Using reference conversations specific to each organization and situation, the AI ​​conference participant can be provided as a participatory AI equipped with conversational skills for multi-party meetings. Accordingly, meeting participants can engage in conversations with the participatory AI according to the organization and situation without disrupting the flow of conversation during the meeting.

[0050] The effects of the present disclosure are not limited to the effects described above, and should be understood to include all effects that can be inferred from the detailed description of the present disclosure or the composition of the invention described in the claims.

[0051] FIG. 1 is a block diagram illustrating a conference service system according to one embodiment.

[0052] Figure 2 is a block diagram illustrating a database according to one embodiment.

[0053] Figure 3 schematically illustrates a block diagram of a conference service device according to one embodiment.

[0054] Figure 4 is a diagram showing the modularization of software implemented by the conference service device illustrated in Figure 3.

[0055] Figure 5 schematically illustrates a block diagram of an online conference room according to one embodiment.

[0056] FIG. 6 illustrates a conceptual diagram illustrating a process by which an online conference room including an agent according to one embodiment is provided to conference participants.

[0057] FIG. 7 is an exemplary diagram illustrating a process for selecting a target conversation from among conversations of meeting participants according to one embodiment.

[0058] Figure 8 is a block diagram illustrating a language model according to one embodiment.

[0059] FIG. 9 illustrates an exemplary flowchart for an agent according to a first embodiment to obtain output data for a target conversation using local internal data.

[0060] FIG. 10 illustrates an exemplary flowchart for an agent according to a second embodiment to obtain output data for a target conversation using specialized field data.

[0061] FIG. 11 illustrates an exemplary flowchart for an agent according to a third embodiment to obtain output data for a target conversation using public data.

[0062] FIG. 12 is a diagram illustrating an example of a conversation response being displayed on a screen of a participant terminal according to one embodiment.

[0063] Figure 13 illustrates an exemplary flowchart in which pre-learning is performed by an additional learning unit according to one embodiment.

[0064] FIG. 14 illustrates an exemplary flowchart of a computer-implemented method performed by a conference service device according to one embodiment.

[0065]

[0066] The advantages and features of the present invention, and the methods for achieving them, will become clearer with reference to the embodiments described in detail below together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided solely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined solely by the scope of the claims.

[0067] When describing embodiments of the present invention, detailed descriptions of known functions or configurations will be omitted if they are deemed to unnecessarily obscure the gist of the invention. Furthermore, the terms described below are defined in light of their functions in the embodiments of the present invention and may vary depending on the intent or custom of the user or operator. Therefore, their definitions should be based on the overall content of this specification.

[0068] When a part of the specification is said to "include" a component, this does not exclude other components, but rather implies the inclusion of other components, unless otherwise specifically stated. Furthermore, terms such as "part" and "module" used in the specification mean a unit that processes at least one function or operation, which may be implemented in hardware, software, or a combination of hardware and software.

[0069] Throughout the specification, “providing” can be interpreted to encompass the process by which a subject acquires specific information or transmits or receives it directly or indirectly to a specific subject, and the performance of related actions required in this process.

[0070] Below, embodiments of the present disclosure are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present disclosure. However, the present disclosure may be implemented in various different forms and is not limited to the embodiments described herein.

[0071] FIG. 1 is a block diagram illustrating a conference service system (1000) according to one embodiment.

[0072] Referring to FIG. 1, a conference service system (1000) may include a conference service device (100), a host terminal (200), one or more participant terminals (300), and a database (400).

[0073] The conference service device (100) can provide a conference environment for users to conduct a conference, and can enable the conference to proceed based on the participation of conference participants. In one embodiment, the conference service device (100) can provide conference participants with a conference environment for users to conduct a conference online in real time or non-real time, and can control the conference to proceed based on the real-time or non-real-time participation of conference participants. For example, the conference service device (100) can create an online conference room based on a conference opening request received from a host terminal (200), and can enable the conference to proceed by providing the online conference room in response to the requests of each participant terminal (300) during the conference period.

[0074] The conference service device (100) may be implemented to perform technical features according to embodiments of the present disclosure. According to one embodiment, the conference service device (100) may be implemented to perform a method for providing an AI-based conference service. For example, during the conference process described above, the conference service device (100) may use an AI-based language model to obtain conversational responses to conversations between conference participants, and provide the obtained conversational responses to the conference participants through the corresponding online conference room. This will be described in detail below with reference to FIG. 2.

[0075] A conference service device (100) according to one embodiment may be implemented as a computer that operates through a computer program to realize the functions described herein. The conference service device (100) may include any type of server and / or any type of user terminal. The server may include any type of computing system or computing device, such as a microprocessor, a mainframe computer, a digital processor, a portable device, and a device controller. The user terminal may include any type of terminal capable of interacting with the server or other computing devices. The user terminal may include, for example, a mobile phone, a smart phone, a desktop computer, a laptop computer, a personal digital assistant (PDA), a slate PC, a tablet PC, and an ultrabook.

[0076] The host terminal (200) refers to a user terminal associated with a conference host. For example, the host terminal (200) may be a terminal corresponding to the user account of the conference host among users using the conference service provided by the conference service device (100). Here, the conference host refers to a user account requesting the opening of a conference, and may correspond to, for example, the user account of the user terminal that transmitted the conference opening request to the conference service device (100), but is not limited thereto. In one embodiment, there may be one or more conference hosts, and may include, for example, a user account requesting the opening of a conference and a user account designated as the conference host by the corresponding user account.

[0077] A participant terminal (300) represents a user terminal associated with a conference participant. For example, the participant terminal (300) may be a terminal corresponding to a user account of a conference participant among users utilizing the conference service provided by the conference service device (100). Here, the conference participant refers to a user account requested to participate in a conference by the conference host, and may correspond to, for example, any one of the user accounts of the conference participants included in the conference opening request of the host terminal (200).

[0078] Hereinafter, for the convenience of explanation, the host terminal (200) and the participant terminal (300) are described separately, but this is not limited thereto. The host terminal (200) and the participant terminal (300) can be understood as distinct concepts depending on whether the user is the conference host or the conference participant, and can be understood collectively as user terminals corresponding to the user account corresponding to the conference host or the conference participant. For example, if there are multiple conferences, the conference host and the conference participants may be different for each conference, and thus the host terminal (200) in one conference may be the participant terminal (300) in another conference, and may also be a user terminal unrelated to another conference.

[0079] According to one embodiment, each of the host terminal (200) and one or more participant terminals (300) can install an application that provides a conference service and store it in memory. For example, a conference service application including a computer program for implementing the functions described herein can be registered in an app store server (not shown) where various applications are uploaded. In addition, each of the host terminal (200) and one or more participant terminals (300) can download and install the conference service application from the application store server, and can communicate with the conference service device (100) through the installed conference service application and receive online conference rooms, conference history, and search results, which will be described later, from the conference service device (100).

[0080] In one embodiment, another conference service application may be a program module capable of communicating with an external device, and such program module may be included in a terminal or another device capable of communicating therewith in the form of an operating system, application module, and other program modules, and may be physically stored on various known memory devices. For example, the conference service application may be an application module that operates on a web basis in an operating system of a computer such as a PC, or may be a mobile application module that operates in an operating system of a smartphone or tablet PC. Meanwhile, such program modules include, but are not limited to, routines, subroutines, programs, objects, components, data structures, etc. that perform technical features according to embodiments of the present disclosure or execute specific data types.

[0081] According to another embodiment, each of the host terminal (200) and one or more participant terminals (300) may receive the following information while communicating with the conference service device (100) on its own without a conference service application, or may receive the following information through a web-based conference service provided by the conference service device (100).

[0082] The database (400) can store information about meetings and various information used to provide conversation responses, as described below. Further details regarding this will be described with reference to FIG. 2.

[0083] FIG. 2 is a block diagram illustrating a database (400) according to one embodiment.

[0084] Referring to FIG. 2, the database (400) may include at least one of local internal data (410), specialized data (420), public data (430), and meeting-related data (440).

[0085] Local internal data (410) may refer to local data and / or internal data whose use of information is restricted within the data storage of the source data within the conference service device (100) or the conference service system (1000). In one embodiment, the local internal data (410) may include local internal data associated with one or more organizations. Here, an organization represents a user group including two or more users, and may include, for example, a corporation, a school, the military, a religious organization, a private organization, and a personal gathering. In one embodiment, the organization may correspond to an organization to which at least some of the user accounts that have subscribed to the conference service provided by the conference service device (100) belong. In one embodiment, the local internal data associated with an organization may include data (e.g., sales data, customer data) created or managed by the organization, for example, internal data to which the organization may at least partially control or have the right to access.

[0086] In one embodiment, local internal data (410) may be categorized by associated organization, and local internal data associated with each organization may be categorized based on one or more types selected from the group consisting of human resources, finance, accounting, legal affairs, management, research, development, production, and sales. For example, each piece of data included in local internal data (410) may be databased in a form that facilitates database search, such as through metadata regarding organizational identifiers, indexing for search, categories, and related keywords.

[0087] In one embodiment, local internal data (410) may be assigned a security level, and for example, each data (e.g., document, segmented document, etc.) included in the local internal data (410) may be assigned one of a plurality of preset security levels. Here, the security level is intended to restrict the use of the local internal data (410), and in one embodiment, it may be in the form of a categorical variable such as an ordinal variable (e.g., level 1 to 5) indicating a high or low security intensity, or a nominal variable indicating a category related to the security intensity (e.g., 1 for executive or higher, 2 for deputy manager or higher but lower than executive, 3 for manager or lower, 4 for outsourced or affiliated companies), but is not limited thereto. The security level according to one embodiment may encompass various forms for directly or indirectly indicating the security intensity, such as a numerical variable calculated through a preset mathematical formula or conditional statement.

[0088] In one embodiment, a security level may be associated with a user's access rights. For example, each of multiple security levels may be configured to have an association with one or more of multiple access rights. Furthermore, the association between security levels and access rights may be managed by organization and configured in various ways, such as through a predetermined operation method or complex conditional statements, based on each organization's security policy.

[0089] In one embodiment, a user's access rights may be set by considering at least one of the following groups: job title, position, document security level, department, employment type, workplace, and affiliate. As described for security levels, a user's access rights may also be set in various ways and in various forms according to each organization's security policy. For example, multiple permission conditions for the aforementioned groups may be set to correspond to multiple access rights, and a permission level corresponding to any one of the permission conditions satisfied by the user information for the aforementioned group may be granted to the user. Alternatively, a permission score may be calculated based on a pre-stored mathematical calculation method in which at least some of the aforementioned groups are applied as variables, and an access right corresponding to a value range within the permission score may be granted to the user.

[0090] Specialization data (420) represents data associated with one or more preset specialized fields. Here, a specialized field represents a field specializing in a specific scope or part of work. For example, the specialized field may include at least one field selected from the group consisting of accounting, law, patents, taxation, medicine, and finance, but is not limited thereto. According to one embodiment, a specialized field may be understood as a concept encompassing a field of work that requires expertise in terms of knowledge, experience, or capabilities in problem solving. Similarly, specialized field data (420) may also be understood as a concept encompassing data available for problem solving that requires expertise in such specialized fields.

[0091] In one embodiment, specialized data (420) may include public data associated with the specialized field. For example, specialized data corresponding to the accounting field may include international accounting standards, national accounting laws, local accounting laws, etc. Depending on the implementation example, specialized data (420) may include data obtained from public data (430) based on categories or keywords associated with the specialized field.

[0092] In another embodiment, specialized data (420) may include non-public data associated with the specialized field. For example, specialized data corresponding to the accounting field may be non-public data with restricted access, documenting specialized knowledge, know-how, etc. provided on a contract basis by a specific company providing accounting consulting services.

[0093] Public data (430) may refer to external data whose use is not restricted from the original data, and may be, for example, data that has been permitted to be disclosed externally by the information provider (e.g., data free from copyright issues). In one embodiment, public data (430) includes data acquired using a preset data collection method (e.g., communication with specific data servers for data collection, data collection through crawling, etc.), and similarly, each data may be databased in a form that is easy to search in the database through indexing for search, metadata regarding related categories and related keywords, etc.

[0094] Meeting-related data (440) includes information about the meeting, and may include, for example, information about an online conference room, which will be described later. In one embodiment, meeting information about the online conference room (e.g., information about meeting participants, the topic of the meeting, the duration of the meeting, etc.) and information obtained through the online conference room (e.g., conversations in a chat room, content, etc.) may be stored and managed as a data set for each meeting. In addition, at least some of the meeting identifier, user identifier, and information about components included in the online conference room may be used as metadata for online conference room information when searching for online conference room information in the database (100).

[0095] In one embodiment, the database (400) may further include a first search engine (not shown) for retrieving data used to provide conversation responses and / or a second search engine (not shown) for retrieving information about online conference rooms. For example, the first search engine may be implemented to provide search results based on a predefined keyword search method (e.g., regular expressions, SQL-based pattern matching, etc.) or a semantic search method (e.g., similarity comparison in vector space, etc.). To this end, each data included in each of the local internal data (410), specialized data (420), and public data (430) may be databased by organization in advance by applying indexing, metadata setting, and encoding methods corresponding to the keyword search method or semantic search method. As another example, the second search engine may be implemented to provide search results for information about online conference rooms based on a keyword search method. To this end, each data included in the conference-related data (440) may be databased by conference by applying indexing, metadata setting, etc. regarding conference information corresponding to the keyword search method.

[0096] According to one embodiment, the database (400) is one of the components of the conference service system (1000), and may be implemented in a form that exists outside the conference service device (100) and can communicate with the conference service device (100). In this case, the database (400) may be managed and controlled by a different external server from the conference service device (100).

[0097] According to another embodiment, the database (400) may be one of the components of the conference service device (100). For example, the conference service device (100) may include a database (400) corresponding to an entity that performs data storage and management, and the database (400) may be included in the conference service device (100) or may exist under the management of the conference service device (100).

[0098] According to another embodiment, the database (400) may be implemented at least partially in the cloud or may include a separate storage server to provide storage space to the conference service device (100).

[0099] At least some of the components included in the conference service system (1000) may be interconnected. For example, the conference service device (100) may be connected to at least one of the host terminal (200) and one or more participant terminals (300) via a network. Here, the network may be configured through various communication networks such as wired and wireless, and may be configured as various communication networks such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). As another example, the conference service device (100) may be connected to the database (400) via a network or electrically.

[0100] Meanwhile, those skilled in the art will appreciate that, in addition to the components illustrated in FIG. 1, the conference service system (1000) may further include other general-purpose components. For example, the conference service system (1000) may further include an internal data provision server (not illustrated) that serves as a collection source for local internal data (410), a mobile carrier server (not illustrated) for providing messages for conference invitations, or a payment server (not illustrated) for payment of usage fees. Alternatively, according to other embodiments, some of the components illustrated in FIG. 1 may be omitted.

[0101] Figure 3 schematically illustrates a block diagram of a conference service device (100) according to one embodiment.

[0102] Referring to FIG. 3, the conference service device (100) may include a memory (110), a communication unit (120), and a processor (130). The configuration of the conference service device (100) illustrated in FIG. 3 is merely a simplified example. In one embodiment, the conference service device (100) may include other components for performing the computing environment of the conference service device (100), and only some of the disclosed components may constitute the conference service device (100).

[0103] The conference service device (100) in the present disclosure may include any type of user terminal and / or any type of server. The user terminal may include any type of terminal capable of interacting with the server or other computing devices. The user terminal may include, for example, a mobile phone, a smart phone, a laptop computer, a personal digital assistant (PDA), a slate PC, a tablet PC, and an ultrabook. The server may include, for example, any type of computing system or computing device, such as a microprocessor, a mainframe computer, a digital processor, a portable device, and a device controller.

[0104] The memory (110) may store at least one instruction that may be executed by the processor (130). The instruction, when executed by the processor (130), may cause the processor (130) to perform the features described throughout the specification. In one embodiment, the memory (110) may store any form of information generated or determined by the processor (130) and any form of information received by the conference service device (100). In one embodiment, the memory (110) may be a storage medium that stores a computer program that causes the processor (130) to perform operations according to embodiments of the present disclosure. Accordingly, the memory (110) may refer to computer-readable media for storing software codes necessary for performing embodiments of the present disclosure, data that is the target of execution of the codes, and execution results of the codes.

[0105] In one embodiment, the memory (110) may mean any type of storage medium. For example, the memory (110) may include at least one type of storage medium among a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (e.g., an SD or XD memory, etc.), a RAM (Random Access Memory), a SRAM (Static Random Access Memory), a ROM (Read-Only Memory), an EEPROM (Electrically Erasable Programmable Read-Only Memory), a PROM (Programmable Read-Only Memory), a magnetic memory, a magnetic disk, and an optical disk. The conference service device (100) may also operate in relation to web storage that performs the storage function of the memory (110) on the Internet. The description of the memory described above is merely an example, and the memory (110) in the present disclosure is not limited thereto.

[0106] The communication unit (120) may be configured regardless of the communication mode, such as wired or wireless, and may be configured with various communication networks, such as a personal area network (PAN) and a wide area network (WAN). In addition, the communication unit (120) may operate based on the known World Wide Web, and may also utilize wireless transmission technologies used for short-distance communication, such as infrared (IrDA: Infrared Data Association) or Bluetooth. For example, the communication unit (120) may be responsible for transmitting and receiving data required to perform a technique according to an embodiment of the present disclosure.

[0107] The processor (130) can control the overall operation of the conference service device (100) and perform a series of operations for users to conduct a conference online. In one embodiment, the processor (130) can be configured with at least one core and can include a processor for data analysis and / or processing, such as a central processing unit (CPU), a graphics processing unit (GPU), a micro controller unit (MCU), a general purpose graphics processing unit (GPGPU), or a tensor processing unit (TPU) of the conference service device (100).

[0108] The processor (130) may be configured to execute at least one instruction stored in the memory (110), and read the computer program stored in the memory (110) to provide a conversation response to a conversation of a conference participant using a language model according to one embodiment of the present disclosure. Here, the language model represents an artificial intelligence-based model trained to generate a natural language response to a natural language input using natural language. According to one embodiment, the language model includes a natural language processing (NLP) model, and may include, for example, a generative pre-trained transformer (GPT) or a bidirectional encoder representations from transformers (BERT). In one embodiment, the language model includes a generative artificial intelligence (AI) model, and may be, for example, a generative AI model trained using at least one selected from the group consisting of a large language model (LLM), a generative adversarial network (GAN), a variational auto-encoder (VAE), and a transformer. For example, a language model can be trained to learn human language using large amounts of text data, perform various tasks based on that language, and generate natural language output appropriate for a given task. The basic structure and learning methods of such language models will be described later.

[0109] A processor (130) according to one embodiment can perform operations for learning a neural network. The processor (130) can perform calculations for learning a neural network, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating weights of a neural network using backpropagation. At least one of a CPU, a GPGPU, and a TPU of the processor (130) can process learning of a network function. For example, a CPU and a GPGPU can together process learning of a network function and classifying data using a network function. In addition, in one embodiment of the present disclosure, processors of a plurality of computing devices can be used together to process learning of a network function and classifying data using a network function. In addition, a computer program executed in a computing device according to one embodiment of the present disclosure can be a CPU, a GPGPU, or a TPU executable program.

[0110] The processor (130) can perform technical features according to embodiments of the present disclosure by executing at least one instruction stored in the memory (110). Various technical features performed by the processor (130) will be described with reference to FIG. 4.

[0111] FIG. 4 is a diagram showing the software implemented by the conference service device (100) illustrated in FIG. 3 in a modularized form. Referring to FIG. 4, the software implemented when the hardware processor (130) executes at least one command stored in the memory (110) may be modularized into at least one of the conference room provider (111), the agent (112), the language model (113), the additional learning unit (114), the security unit (115), and the learning unit (116). For example, each of the conference room provider (111), the agent (112), the language model (113), the additional learning unit (114), the security unit (115), and the learning unit (116) may be implemented as a computer program, and commands and data for each execution may be stored in the memory (110) and executed by the processor (130), but are not limited thereto.

[0112] According to one embodiment of the present disclosure, the conference room provider (111) may be implemented to provide an online conference room to conference participants. Specifically, the conference room provider (111) may provide an online conference room (10) to at least one participant terminal (300) among the conference participants participating in the online conference room via a network.

[0113] Here, the online conference room (10) corresponds to a conference and can be understood as a concept including a data space allocated for storing and managing data related to the conference, but is not limited thereto. The online conference room (10) according to one embodiment can be understood as a concept encompassing an online conference space or a conference environment therefor provided to conference participants using data stored in the aforementioned data space. Accordingly, providing the online conference room (10) can include, for example, a process of providing data included in the online conference room (10) to participant terminals (300), and, for another example, can be understood as encompassing a series of processes of connecting participant terminals (300) through a virtual conference space corresponding to the online conference room (10) to provide a conference environment for conference participants to conduct a conference using the corresponding data.

[0114] Figure 5 schematically illustrates a block diagram of an online conference room (10) according to one embodiment.

[0115] Referring to FIG. 5, the online conference room (10) may include at least one selected from the group consisting of a basic information object (11), a content object (12), a conversation thread object (13), and a meeting history object (14). Here, the object refers to a processing unit of each component included in the online conference room (10). Each object according to one embodiment corresponds to a logical storage unit in which the corresponding data is stored, and may be allocated to a different storage space within the online conference room (10), but is not limited thereto. For example, although each object is one of the components of the online conference room (10), an authorized user may independently process a query, deletion, input, or modification, etc., through one of the components of another online conference room (10). Meanwhile, each of the basic information object (11), the content object (12), the conversation thread object (13), and the meeting history object (14) according to one embodiment may be used interchangeably with basic information, content, conversation thread, and meeting history, respectively.

[0116] The basic information object (11) may include meeting information. In one embodiment, the meeting information includes at least one of the meeting period, meeting content, information about the meeting host, information about the meeting participants, and meeting classification information, and may further include a meeting title, meeting objectives, etc. In one embodiment, the basic information object (11) may include meeting information obtained from a meeting opening request received from the host terminal (200), and may further include information obtained during the meeting process (e.g., announcements, reference materials, meeting result reports, etc.).

[0117] The content object (12) may include video data. Here, the video data is video data related to the meeting content, and refers to a video explaining the content to be covered at the meeting, but is not limited thereto. For example, the video data may be a video shot of a meeting host explaining a presentation material regarding the meeting agenda, or a working video virtually simulating technical content corresponding to the meeting topic. Alternatively, the video data may be a reference video regarding a topic that may be covered at the meeting, or a video or an access link to the video for ideation or icebreaking that is not directly related to the meeting content. In one embodiment, the content object (12) includes video data acquired from the host terminal (200) or another user's terminal (not shown) designated by the host terminal (200), and may further include additional video data acquired thereafter.

[0118] A conversation thread object (13) may include one or more conversation threads. Here, a conversation thread is a thread of conversation units and may include multiple conversations by conference participants. In one embodiment, a conversation thread may be understood as a concept encompassing a workspace for sequentially processing multiple conversations generated by conference participants, and may be provided in the form of a messenger-based chat room using a message queue and a handler, for example. In addition, a conversation refers to a processing unit in a thread, and in one embodiment, may be based on at least one input of text, image, video, and audio from conference participants. For example, each conversation may be a message based on text input, or a message based on image, video, or audio input.

[0119] A meeting history object (14) may represent a historical relationship between a corresponding online meeting room (10) and one or more other online meeting rooms (10). The relationship may include a relationship and / or a correlation between meetings, and in one embodiment, may include information about past meetings whose meeting information (e.g., meeting content, meeting topic, technical field, etc.) is similar to a preset level or higher, and past meetings that have a predetermined correlation with the corresponding meeting (e.g., direct connection meeting, indirect connection meeting, derived meeting, etc.). To this end, the relationship may be analyzed by the meeting room provider (111) from a database (400) storing a plurality of online meeting rooms (10), and a meeting history object (14) including information about the relationship may be created or updated as a result of the analysis.

[0120] In one embodiment, the meeting history object (14) may include a visualization of the chronological and / or hierarchical relationships of meetings prior to the creation of the online meeting room (10) based on the meeting classification information to be described later. For example, the meeting history object (14) may include the chronological and / or hierarchical relationships of meetings based on at least one of the meeting host, meeting participant, department name, project name, specific work or task in the workflow, technical field, and past meetings performed prior to the meeting, which are included in the meeting classification information to be described later. For example, the meeting history object (14) may include at least one of (i) a history diagram that visualizes the relationships described above, (ii) a meeting results report summarizing the meeting results of one or more other online meeting rooms (10) associated with the online meeting room (10), and (iii) an access link to one or more other online meeting rooms (10) associated with the online meeting room (10). For example, a history diagram may include an indication of a chronological relationship (e.g., arrangement order) between a current meeting and past meetings, an indication of a connection relationship (e.g., direct connection, indirect connection, derived connection, etc.) (e.g., solid line, dotted line), and an indication of a current meeting (e.g., highlight), and each past meeting may be mapped with an access link to an online conference room (10) corresponding to each past meeting. Meanwhile, a means for expressing these relationships may utilize various general-purpose expression methods such as a flowchart, a structure diagram, a block diagram, or a multi-dimensional diagram (e.g., an image, a graph, a matrix), and is not limited to the forms exemplified above.

[0121] The conference room provider (111) can provide an online conference room (10) to each participant terminal (300) of the conference participants via a network. As described above, providing the online conference room (10) can be understood as a concept including providing information included in the online conference room (10) and / or providing a conference environment for users to utilize the information by connecting users through a virtual conference space. For example, the conference room provider (111) can provide conference information through a basic information object (11), provide a multimedia environment in which each conference participant can watch a video whenever they want through a content object (12), provide a conversation environment (e.g., a chat room) in which conference participants can exchange opinions through a conversation thread object (13), and provide history information on the relationship between other past conferences related to the current conference through a conference history object (14).

[0122] In one embodiment, the conference room provider (111) can provide an online conference room (10) through a network in response to a request from each participant terminal (300) of the conference participants during the conference period, thereby allowing the conference to proceed through non-real-time participation of the conference participants. For example, the conference room provider (111) can create a conference session corresponding to the online conference room (10) and in which the participant terminals (300) can participate non-real-time during the conference period, and can allow the participant terminals (300) of the conference participants to be logically connected through the session during the conference period. The host terminal (200) or each participant terminal (300) can request access to the online conference room (10) to the conference room provider (111) through a conference service application at any time during the conference period, and the conference room provider (111) can control the non-real-time conference to proceed by providing the online conference room (10) to the corresponding terminal through the corresponding session whenever a connection is requested.

[0123] The provision of the above-described online conference room (10) can be performed regardless of the current location of the participant terminal (300), the connection time, or whether other conference participants are currently connected. For example, even if each participant terminal (300) is located remotely from the other, requests connection at a late hour, or even if no user is connected to the online conference room (10), the online conference room (10) can be provided from the conference room provider (111) in response to an unrelated connection request. Accordingly, the conference can proceed through non-real-time participation of conference participants. The conference host and conference participants can connect to the online conference room (10) at a desired time and location, watch the video, or asynchronously exchange opinions on the conference agenda presented in the video in a chat room to participate in the conference in non-real time. In addition, each conference participant can participate in the conference in non-real time at a desired time, while communicating in real time with other conference participants who are simultaneously connected to the online conference room (10) through a chat room.

[0124] In another embodiment, the conference room provider (111) may provide an online conference room (10) to each participant terminal (300) of the conference participants via a network in response to the arrival of the conference start time or a conference start request from the host terminal (200), thereby allowing the conference to proceed through real-time participation of the conference participants. For example, the conference room provider (111) may create a conference session corresponding to the online conference room (10) and in which the participant terminals (300) can participate in real-time, and may logically connect the participant terminals (300) of the conference participants through the session in response to the conference start time requested by the host terminal (200). The conference room provider (111) provides the online conference room (10) to the participant terminal (300) that approves the connection through the session, and the host terminal (200) or each participant terminal (300) can participate in the conference in real time by using video input, voice input, chat input, etc. while the conference session is maintained, and the conference room provider (111) can control the progress of the conference in real time by terminating the session in response to a request for terminating the conference by the host terminal (200).

[0125] During the conference process described above, the conference room provider (111) can receive conversations from conference participants participating in the online conference room (10). According to one embodiment, the conversations may be received from the host terminal (200) or one or more participant terminals (300) based on an online conference environment, or may be received from the host terminal (200) based on an online conference environment combined with an offline conference. For example, the conference service device (100) is connected in real time or non-real time with the host terminal (200) and the first participant terminal (310), the second participant terminal (320) to the Nth participant terminal (N is a natural number), and the conference room provider (111) can receive conversations based on an input method of text, image, video, or audio from each of the participant terminals (300) through the online conference room (10). As another example, the conference service device (100) is connected to the host terminal (200) based on a real-time offline conference environment, and the conference room provider (111) can receive a conversation based on an audio input method by either the conference host or conference participants in the same space from the host terminal (200) through the online conference room (10). Alternatively, conversations may be received in a mixed form of the above-described examples.

[0126] The conference room provider (111) can update the online conference room (10) to include conversations received from conference participants, and for example, can process each received conversation in the form of text, image, video, or audio and add it to the conversation thread object (13). In one embodiment, the conference room provider (111) can convert the received conversation into text based on an audio input method and add it to the conversation thread object (13). Specifically, the conference room provider (111) can identify two or more conference participants from among the conversations received based on an audio input method, and process each of the received conversations based on the audio input method as a conversation by a conference participant corresponding to each conversation among the two or more conference participants. In addition, the conference room provider (111) can convert the received conversations into text based on an audio input method, process the converted conversations as messages by the conference participants corresponding to each conversation, and add them to the conversation thread object (13).

[0127] The conference room provider (111) may, during the process of providing an online conference room (10), provide an agent (112) connected to a language model. Here, the agent (112) represents a software module that functions as a virtual conference participant participating in the online conference room (10) and as an AI (Artificial Intelligence)-based conference participant. The agent (112) may be implemented to perform technical features according to embodiments for conference participation described throughout the specification.

[0128] An agent (112) according to one embodiment may be understood as a software agent capable of performing tasks for functioning as a conference participant in a conference environment corresponding to an online conference room (10). For example, the agent (112) may be located in an online conference room (10) corresponding to a virtual conference space and may function as an AI (Artificial Intelligence) conference participant participating in the online conference room (10) together with conference participants. In one embodiment, the agent (112) may be a conference participant object participating in the online conference room (10) together with conference participants. For example, the agent (112) may be understood as an object having properties and functions corresponding to a virtual conference participant without a physical entity, and agents (112) participating in each of a plurality of online conference rooms (10) may be identified from each other by their respective identifiers.

[0129] According to one embodiment, the agent (112) may be implemented in the form of an intelligent agent designed to initiate participation in the online conference room (10) as a conference participant by being granted the conference participant's authority for the online conference room (10) by the conference room provider (111), and autonomously perform tasks for conference participation while interacting with conference participants without user intervention. According to another embodiment, the agent (112) may be implemented in the form of a service program that can be bound and executed with the conference room provider (111) during the process of providing the online conference room (10). However, the description of the agent described above is merely an example, and the agent (112) in the present disclosure is not limited thereto, and one or more modules that perform at least some of the technical features described throughout the specification may be implemented in an integrated form or in various other forms.

[0130] In one embodiment, the conference room provider (111) may determine whether an agent (112) participates in an online conference room (10) based on a conference opening request. For example, when creating an online conference room (10) based on a conference opening request from a host terminal (200), if the conference opening request includes an invitation request for an AI conference participant, the conference room provider (111) may grant the conference participant's authority for the online conference room (10) to the agent (112) before the conference starts. In another embodiment, when an invitation request for an AI conference participant is received from the host terminal (200) or the participant terminal (300) while providing the online conference room (10), the conference room provider (111) may grant the conference participant's authority for the online conference room (10) to the agent (112) during the conference.

[0131] In one embodiment, the conference room provider (111) can allow an agent (112) to participate in the online conference room (10) by associating the agent (112) with the online conference room (10) and controlling the agent (112) to operate in association with the online conference room (10). In one embodiment, the conference room provider (111) can allow an agent (112) to participate in the online conference room (10) by including an agent (112) associated with a language model in the online conference room (10) and providing the online conference room (10) including the agent (112). In one embodiment, including an agent (112) in an online conference room (10) may include a series of processes for creating a logical connection structure between the agent (112) and the online conference room (10) and for allowing the agent (112) to operate as a conference participant of the online conference room (10) based on the structure, and for example, may be interpreted as encompassing execution control, environment setting, and operation control for the agent (112) for allowing the agent (112) to function as a virtual conference participant while being located in a virtual conference space corresponding to the online conference room (10).

[0132] FIG. 6 illustrates a conceptual diagram for explaining a process in which an online conference room (10) including an agent (112) according to one embodiment is provided to conference participants.

[0133] Referring to FIG. 6, the conference room provider (111) may create a first online conference room (10a) based on a conference opening request from a first conference host for a first conference, and include a first agent (112a) corresponding to an AI conference participant of the first conference in the first online conference room (10a). The conference room provider (111) may retrieve the first agent (112a) for the first conference from an agent (112) stored in the memory (110), and perform a series of processes of setting and executing the first agent (112a) so that the first agent (112a) is located in a virtual space corresponding to the first online conference room (10a) and interacts with conference participants participating in the first online conference room (10a) to participate in the conference. When a first online conference room (10a) including a first agent (112a) is prepared, the conference room provider (111) can create a first session corresponding to the first online conference room (10a) and provide the first online conference room (10a) including the first agent (112a) to the participant terminals (300) of the corresponding conference participants (e.g., Users #1) in real time or non-real time through the first session during the conference period. While the first conference is in progress, the first agent (112a) can perform tasks to function as a conference participant participating in the conversation of the conference participants through the chat room of the first online conference room (10a). In a similar manner, the conference room provider (111) can provide a second online conference room (10b) including a second agent (112b) to the participant terminals (300) of the corresponding conference participants (e.g., Users #2) through a second session during the conference period.

[0134] In one embodiment, the agent (112) may be activated for the duration of a meeting in the online conference room (10). For example, the agent (112) may be activated while a session corresponding to the online conference room (10) is maintained, and may be deactivated when the session ends.

[0135] In one embodiment, the agent (112) may perform operations to function as a conference participant object of the online conference room (10) using a language model (113). As described above, the language model (113) is a model trained to generate natural language responses to natural language input using natural language, and may be, for example, a generative AI model. The language model (113) according to one embodiment may be included in the conference service device (100), or may communicate with the conference service device (200) via a network as a separate artificial intelligence model. In one embodiment, the language model (113) may be stored in the memory (110), executed by the conference room provider (111), and communicate with the agent (112). For example, a language model (113) learned by a conference service device (100) is loaded from a memory (110) and executed, and each of a first agent (112a) and a second agent (112b) can communicate with the language model (113) when use of the language model (113) is required while performing tasks for functioning as a conference participant. In another embodiment, the language model is stored in an external device (not shown) (e.g., a language model provider) connected to the conference service device (100) via a network, and can communicate with the agent (112). For example, a language model learned by an external device is prepared, and each of a first agent (112a) and a second agent (112b) can communicate with the language model (e.g., using an API (Application Programming Interface)) through the external device when use of the language model is required. Hereinafter, embodiments in which an agent (112) utilizes a language model (113) are described, but it should be understood that embodiments in which the agent (112) communicates with a language model as a separate artificial intelligence model and performs actions for meeting participation can also be implemented in a similar manner. Meanwhile, specific embodiments in which the agent (112) utilizes a language model (113) will be described later in each section describing the relevant actions.

[0136] According to one embodiment of the present disclosure, the agent (112) may be implemented to acquire conversation(s) received from conference participants participating in an online conference room (10). During the conference, the agent (112) may acquire at least one of the conversations among the conference participants received by the conference room provider (111) as a conversation for conference participation. For convenience of explanation, the term "conference participant" may be used herein to encompass the conference host.

[0137] In one embodiment, the agent (112) may be granted the authority to input and receive conversations on a par with conference participants, and may acquire conversation(s) received from conference participants based on the granted authority. Specifically, the agent (112) may be granted the authority of a conference participant, including the authority to input and receive conversations on a par with conference participants, from the conference room provider (111) during the process of being included in the online conference room (10), and may access the conversation thread object (13) of the online conference room (10) to which conversations of conference participants entered into the chat room are added based on the authority. For example, the agent (112) may receive conversations processed in the conversation thread object (13) in real time based on the authority to receive conversations, or may receive the most recently included conversation among the conversations, a predetermined number of recent conversations, or all conversations corresponding to the time of conversation processing.

[0138] According to one embodiment of the present disclosure, the agent (112) may be implemented to select a target conversation from among the acquired conversation(s). Here, the target conversation represents a conversation requiring a response. Subsequent processes may be performed to provide a conversation response for the conversation selected as the target conversation.

[0139] In one embodiment, the agent (112) can acquire conversations received from conference participants and use the conversations to determine whether the conversations correspond to target conversations. For example, the agent (112) can determine each conversation received from conference participants in real-time or non-real-time as a target conversation if the conversation is analyzed as requiring a response.

[0140] In another embodiment, the agent (112) may acquire conversations received from conference participants and use the conversations to select a target conversation among them. For example, when a conversation is received from a conference participant, the agent (112) may acquire multiple conversations including the conversation from the conversation thread object (13), determine whether any of the acquired conversations require a response, and if it is determined that a conversation requires a response, determine the conversation as the target conversation.

[0141] FIG. 7 is an exemplary diagram illustrating a process for selecting a target conversation from among conversations of meeting participants according to one embodiment.

[0142] Referring to FIG. 7, the agent (112) can select a target conversation requiring a response from among the first message (21) and the second message (22) included in the acquired conversations (20). For example, if the first message (21) corresponds to a reply, opinion, or other type, and the second message (22) corresponds to a question, request, or command, the agent (112) can select only the second message (22) from among the first message (21) and the second message (22) as the target conversation. For example, the type of each conversation may be managed to correspond to the conversation type received together when each conversation is received, or may be determined as a result of text analysis (e.g., text structuring, text classification, extraction, etc.), syntax analysis (e.g., contextual understanding), or conditional analysis (e.g., inclusion of question marks, etc.) for each conversation.

[0143] In one embodiment, target conversation selection can be performed by having the agent (112) select target conversations from among conversations in a rule-based manner. Specifically, the agent (112) can select target conversations by analyzing whether each conversation (or conversations) conforms to a predetermined selection rule.

[0144] In one embodiment, the selection rule may be based on at least one selected from the group consisting of the type of conversation, the intent of the conversation, whether the other meeting participant is nominated, whether the AI ​​meeting participant is nominated, the context of the conversations, the time interval between the conversations, and the number of conversations during a given time interval. For example, if the type of conversation analyzed from each conversation falls into a preset response-required type (e.g., question, request, command, etc.), if the intent of the conversation detected from each conversation corresponds to a preset response-required intent (e.g., action suggestion, information confirmation, etc.), if each conversation includes a preset text corresponding to a call from an AI conference participant (e.g., AI participant name), if the context analyzed from the conversations is analyzed to be a context requiring a response when considering the preceding and following conversations of the conference participant in question or the preceding and following conversations of any conference participant together, if the time interval between recent conversations among the conversations is greater than a predetermined reference value and thus the conversation update speed is relatively slow, if the number of conversations during the recent predetermined time is less than a predetermined reference number and thus the conversation activity level is relatively low, or if a preset condition based on at least some of these is satisfied, the agent (112) may determine that the conversation corresponds to a target conversation. For another example, if the aforementioned exemplary cases do not apply, or if the conversation includes text corresponding to at least one of the meeting participants (e.g., participant name, position, etc.), the agent (112) may determine that the conversation does not correspond to the target conversation.

[0145] According to another embodiment, the selection of target conversations may be performed by having the agent (112) select target conversations using a language model (113) trained to select target conversations or a target conversation selection model, which is a separate artificial intelligence model. Specifically, the agent (112) executes the language model (113) (or target conversation selection model), transfers the acquired conversation(s) to the language model (113) (or target conversation selection model), and the language model (113) (or target conversation selection model) can select target conversations from among the transferred conversations.

[0146] In another embodiment, the language model (113) may be a generative AI model that is pre-trained to generate natural language responses to natural language input using a large amount of text data, and further trained (e.g., fine-tuned) to select conversations requiring a response from the conversations based on supervised learning using multiple training datasets. Each training dataset may include training input data including multiple conversations and training answer data indicating whether a response is required. For example, in the process of fine-tuning, the pre-trained generative AI model may be modified to suit the task of selecting target conversations, and the error may be calculated by comparing the output of the pre-trained model for the training input data with the corresponding training answer data, and the parameters of the model may be updated according to backpropagation to reduce the error.

[0147] In another embodiment, the target dialogue selection model may be a natural language processing-based model trained to select dialogues requiring a response based on multiple dialogues and the need for a response. For example, the target dialogue selection model may be trained to select dialogues requiring a response from dialogues through supervised learning using the aforementioned multiple training datasets, and may be stored in memory (110).

[0148] The selection of target conversations described above, according to one embodiment, has a significant feature in that it allows AI meeting participants to fluidly participate in online conference chat rooms where multiple participants exchange opinions, questions, and answers regarding meeting agenda items, by selecting conversations requiring responses without disrupting the flow of conversation. For example, if AI were to intervene in every conversation, as with conventional chatbot technology, the meeting would not proceed smoothly. Alternatively, if AI only responded to questions or commands directed at it, it would be limited to a supporting role in the meeting. However, the selection of target conversations described above, according to one embodiment, allows conversation responses to be provided for only those conversations requiring responses, rather than all conversations. Accordingly, the AI ​​meeting participant can function as a fully functioning meeting participant, autonomously responding to conversations requiring responses even if the meeting participant does not designate the AI.

[0149] According to one embodiment of the present disclosure, the agent (112) may provide a target conversation as input data to the language model (113) and obtain output data for the input data from the language model (113). Specifically, the agent (112) may provide input data including the target conversation to the language model (113), so that the language model (113) may generate output data for the target conversation using local internal data (410). The process of providing input data according to one embodiment may further include a preprocessing process for converting the input data into a form suitable for use as input in the language model (113), and may further include a process for text preprocessing such as tokenization, cleaning, encoding, and embedding of data. These processes may be performed with reference to known technologies, and a description thereof will be omitted.

[0150] In one embodiment, the agent (112) can provide the selected target conversation as input data to the language model (113) and obtain output data from the language model (113). For example, as described above, the language model (113) is a generative AI model trained to generate natural language responses to natural language inputs, the agent (112) inputs the selected target conversation to the language model (113), and the language model (113) generates output data (e.g., our company's operating profit in 2023 is OOO) for the input target conversation (e.g., what was our company's operating profit last year?), and the agent (112) can obtain the output data for the corresponding target conversation.

[0151] In one embodiment, the language model (113) may generate a natural language response to a natural language input using at least one of local internal data (410), specialized data (420), and public data (430). For example, the language model (113) is a generative AI model trained to generate a natural language response to a natural language input, and the agent (112) may request the language model (113) to generate output data for a target conversation using at least one of local internal data (410), specialized data (420), and public data (430). In another example, the language model (113) is a generative AI model trained to generate a natural language response to a natural language input using at least one of local internal data (410), specialized data (420), and public data (430), and the agent (112) may provide the target conversation to the language model (113) to obtain output data for the target conversation from the language model (113).

[0152] Below, various embodiments in which the above-described operations are performed will be described in more detail with reference to FIGS. 8 to 11.

[0153] Figure 8 is a block diagram illustrating a language model (113) according to one embodiment.

[0154] Referring to FIG. 8, the language model (113) may include at least one selected from the group consisting of one or more Colleague models (1131), Expert models (1132), and Participant models (1133).

[0155] According to one embodiment, the Colleague model (1131) may be trained to generate natural language responses to natural language input using local internal data (410). For example, the Colleague model (1131) may be a generative AI model trained to converse using local internal data associated with a specific organization among the local internal data (410), and may be a conversational model implemented to enable collegial-level conversations with meeting participants belonging to the organization.

[0156] In one embodiment, the Colleague model (1131) may be obtained by applying additional learning using local internal data (410) to a pre-trained generative AI model that is trained to converse with users based on public data. For example, the Colleague model (1131) may be fine-tuned to generate natural language responses (e.g., answers) to natural language inputs (e.g., questions) based on local internal data (410) by using a plurality of first learning datasets, each of which includes learning input data (e.g., questions about internal information of an organization) and learning answer data (e.g., answers using internal information of an organization) related to the local internal data (410). According to one embodiment, the plurality of first learning datasets may be obtained by performing various tasks for preparing learning data, such as extracting or collecting data from the local internal data (410), converting, processing, and labeling data.

[0157] In one embodiment, the Colleague model (1131) may be trained to generate natural language responses to natural language inputs using local internal data (410) whose security levels at least partially correspond to the access rights granted to each meeting participant among the local internal data (410). In one embodiment, the Colleague model (1131) may use multiple training datasets, each including training input data (e.g., questions about internal information of an organization with a security level, the user's access rights) and training answer data (e.g., answers using internal information of an organization with a security level corresponding to the user's access rights) related to the local internal data (410) with a security level, during the above-described fine-tuning process. For example, the Colleague model (1131) may be trained to generate responses only based on local internal data (410) with a security level corresponding to the user's access rights during the above-described fine-tuning process. For example, different answers may be generated for the same question depending on the user's access rights.

[0158] The Colleague model (1131) according to the above-described embodiments can generate output data as a response using local internal data (410) when a target conversation is input as the above-described learning is performed.

[0159] FIG. 9 illustrates an exemplary flowchart for an agent (112) according to the first embodiment to obtain output data for a target conversation using local internal data (410).

[0160] Referring to FIG. 9, in step S910, the agent (112) can select a first target conversation that includes a conversation requiring a response using local internal data (410) from among the acquired conversation(s).

[0161] In one embodiment, the agent (112) may select a first target conversation from among the conversation(s) based on the selection rules described above, which may further include whether the conversation(s) contain text associated with local internal data (410) (e.g., company name, product name, sales, etc.).

[0162] In another embodiment, the language model (113) (or target conversation selection model) is trained to select a first target conversation, and the agent (112) can select the first target conversation using the language model (113) (or target conversation selection model). For example, the Colleague model (1131) may be trained to select a first target conversation from among conversations and generate output data for the first target conversation using local internal data (410). This training may additionally utilize training datasets that take into account the characteristics of conversations using local internal data (410). For example, the training input data of each training dataset may include a series of conversations (e.g., multiple conversations including questions about internal information of the organization and conversations unrelated to internal information of the organization), and the training answer data may include a label for the target conversation (e.g., a question about internal information of one of the conversations) and a label for the response to the target conversation (e.g., an answer to the question using internal information of the organization).

[0163] In step S920, the agent (112) provides the first target conversation as input data to the language model (113), and in step S930, the agent (112) can obtain output data for the first target conversation generated using local internal data (410) from the language model (113). For example, the agent (112) inputs the first target conversation (e.g., “What was your company’s operating profit last year?”) into the language model (113), and the language model (113) can generate output data for the first target conversation (e.g., “Your company’s operating profit in 2023 is OOO”) based on local internal data (410) (e.g., corporate accounting data) associated with the organizations of meeting participants participating in the online conference room (10).

[0164] In one embodiment, the agent (112) may request the language model (113) to generate output data for the first target conversation by providing input data including the first target conversation to the Colleague model (1131). For example, since the Colleague model (1131) is trained to respond to input using local internal data (410), as the first target conversation is input from the agent (112), the Colleague model (1131) may generate output data for the first target conversation using the local internal data (410).

[0165] In another embodiment, the agent (112) may provide input data including a first target conversation and access rights of conference participants to the Colleague model (1131), and request the Colleague model (1131) to generate output data for the first target conversation using the access rights of the conference participants. For example, the agent (112) may determine a security level corresponding to the access rights of the conference participants or the access rights of the online conference room (10) based on a pre-stored security policy of the organization to obtain output data considering the access rights of the conference participants. The agent (112) may generate a prompt to instruct the Colleague model (1131) to generate output data using local internal data (410) to which the determined security level has been assigned, and input the generated prompt to the Colleague model (1131) to obtain output data generated using the local internal data (410) corresponding to the security level from the Colleague model (1131).

[0166] In one embodiment, a prompt may include specific requests for instructing a language model (113) trained to respond to an input to generate a response using information other than the input and / or the information. In one embodiment, the agent (112) may generate the prompt using a prompt template corresponding to a request for generating a response to a first target conversation using a security level among a plurality of pre-stored prompt templates. For example, the prompt may include conditions regarding the security level of local internal data (410) used as a basis for generating the response, a response method in case of a response that exceeds the security level (e.g., no response, notification that access is not authorized, etc.), etc. The prompt generation may utilize various algorithms known for prompting operations.

[0167] In another embodiment, the agent (112) may provide input data including a first target conversation and local internal data (410) associated with the first target conversation to the language model (113), and request the language model (113) to generate output data for the first target conversation using the local internal data (410). In this embodiment, the language model (113) may be a conversational model trained to converse with a user using public data, and may be, for example, a general-purpose generative AI model that is capable of communicating with the conference service device (100) as a model provided by an external device.

[0168] Specifically, the agent (112) can obtain a search result for the local internal data (410) associated with the first target conversation among the local internal data (410) stored in the database (400). The search result may include at least one of an identifier, an index, metadata (e.g., classification, summary, etc.), a security level, an author, a data storage path (e.g., a physical storage location or a search key, an address, etc.) of each searched data, a corresponding data file (e.g., a document file), and a preset format of converted data converted from the corresponding data file (e.g., text data extracted from the document file). In one embodiment, the agent (112) can obtain the search result from the database (400) using the keyword search method and / or the semantic search method described above. For the former example, each data (e.g., document, etc.) stored in the database (400) may be managed based on document indexing or metadata for access to and retrieval of the data. The agent (112) can extract one or more keywords from the first target conversation, perform a keyword search on the database (400) based on a keyword search method, and retrieve data from the database (400) that include the keyword or content similar to the keyword at a relatively high frequency as a search result. For example, each data (e.g., document, etc.) stored in the database (400) can be managed based on a semantic vector encoded from texts (e.g., keywords, descriptions, titles, document content, etc.) related to each data for accessing and searching the data. The agent (112) can encode the first target conversation and convert it into a semantic vector, perform a semantic search on the database (400) based on a semantic search method, and retrieve data that have a relatively high semantic similarity to the first target conversation from the database (400) as a search result.

[0169] In addition, the agent (112) may generate a prompt to instruct the language model (113) to generate output data for the first target conversation using the acquired search results, and provide the generated prompt to the language model (113). For example, the agent (112) may generate the prompt using a prompt template corresponding to a request for generating a response to the first target conversation using the search results from among a plurality of prompt templates. For example, the prompt may include search results (e.g., documents, parts of documents), a method of applying the search results (e.g., answers only within the scope of the provided documents, answers using public data together, etc.), a description of the search results (e.g., a description representing the provided documents, etc.), a security policy regarding the use of local internal data (410) (e.g., no learning performed), a recommended format of the output data (e.g., adding a title based on the conversation partner's job title, a listing order of elements included in the output data (e.g., responses, response grounds, etc.), etc.).

[0170] In one embodiment, the acquisition and use of search results may be performed based on Retrieval Augmented Generation (RAG). For example, an agent (112) may generate an input query related to a target conversation and use a search engine in a database (400) to search for local internal data (410) related to the input query. The retrieved information may be converted into vectors in a high-dimensional space, and data with a relatively high similarity to the input query may be selected from the knowledge vectors stored in the database (400). These search results may serve as a reference source for all facts and contexts required by the language model (113).

[0171] In another embodiment, the agent (112) may provide input data including more conversations to the Colleague model (1131) so as to request that the Colleague model generate output data for the first target conversation, including conversations of conference participants not selected for the first target conversation.

[0172] According to the above-described embodiment, an agent (112) corresponding to an AI conference participant can provide a professional response to a first target conversation using local internal data (410) without interrupting the conversation flow of the conference participants. Accordingly, not only can limitations in response reliability, such as hallucinations that frequently occur in generative AI, be overcome, but the AI ​​conference participant can also function as a colleague AI that is well-informed about the local internal data (410) related to the organization to which the conference participant belongs and can converse with them, and can participate in the conference at an equal level with other conference participants.

[0173] In addition, as described above, since the local internal data (410) may include various information held by the organization, a significant level of security is required for the use of the local internal data (410). According to the above-described embodiment, the use of local internal data (410) with a security level that does not correspond to the access rights of each meeting participant is restricted during the response generation process, thereby enhancing the security level. Alternatively, according to another embodiment described above, the language model (113) only utilizes the provided local internal data (410) when generating a response and does not perform learning on the same, thereby enhancing the security level of the local internal data (410).

[0174] Referring back to FIG. 8, an Expert model (1132) according to one embodiment may be trained to generate a natural language response to a natural language input using specialized field data (420) associated with a preset specialized field. For example, the Expert model (1132) is a generative AI model trained to have a conversation using specialized field data associated with a specific specialized field among specialized field data (420), and may be a conversational model implemented to enable a conversation at the level of an expert with expertise in the specialized field in terms of knowledge, experience, or competency.

[0175] In one embodiment, the Expert model (1132) may be obtained by applying additional learning using specialized data (420) to a pre-trained generative AI model that is trained to converse with users based on public data. For example, the Expert model (1132) may be fine-tuned to respond (e.g., answer) in natural language to a natural language input (e.g., question) based on specialized data (420) by using a plurality of second training datasets each including training input data (e.g., questions related to a specific specialized field) and training answer data (e.g., answers applied with knowledge or know-how of the specialized field).

[0176] In one embodiment, the Expert model (1132) may include multiple models trained using specialized field data (420) corresponding to each specialized field, for each of multiple specialized fields. For example, second learning datasets may be prepared for each of multiple preset specialized fields (e.g., accounting, law, patent, tax, medicine, finance, etc.), and multiple Expert models (1132) trained to provide optimized specialized answers based on knowledge or know-how of each specialized field to questions may be prepared using each of the second learning datasets. For example, the agent (112) may select an Expert model (1132) corresponding to one specialized field that is most relevant to the second target conversation among the multiple Expert models (1132), and provide the second target conversation as input data to the selected Expert model (1132), thereby obtaining output data from the corresponding Expert model (1132). For another example, in the process of an agent (112) participating in an online conference room (10), an Expert model (1132) corresponding to one specialized field is selected by the host terminal (200) from among multiple Expert models (1132), and the agent (112) can participate in the online conference room (10) using the selected Expert model (1132).

[0177] According to one embodiment, multiple second learning datasets can be obtained from public data related to the relevant field of expertise. For example, specialized data (420) or data intensively acquired from various external devices within the relevant field of expertise can be used to prepare training data.

[0178] According to another embodiment, a plurality of second learning datasets may be obtained from local internal data (410) or specialized field data (420) associated with the relevant field of expertise. For example, a plurality of second learning datasets associated with the field of accounting may be prepared using the local internal data (410) or specialized field data (420) provided from a device of a company providing accounting consulting services, and an Expert model (1132) associated with the field of accounting trained using the second learning datasets may be prepared. For example, a conference service may be provided in such a way that an agent (112) using an Expert model (1132) associated with a specific specialized field participates in an online conference room (10) in response to an invitation request from a host terminal (200) or a participant terminal (300), and such conference service may also be provided through a paid payment linked to the relevant company depending on the level of specialized knowledge, know-how, difficulty, etc. applied to the learning datasets.

[0179] The Expert model (1132) according to the above-described embodiments can generate output data using specialized field data (420) when a target conversation is input as the above-described learning is performed.

[0180] Figure 10 illustrates an exemplary flowchart for an agent (112) according to a second embodiment to obtain output data for a target conversation using specialized field data (420). Detailed descriptions of embodiments similar to the technical features described above will be omitted.

[0181] Referring to FIG. 10, in step S1010, the agent (112) may select a second target conversation that includes a conversation requiring a response using specialized field data (420) from among the acquired conversation(s).

[0182] In one embodiment, the agent (112) may select a second target conversation from among the conversation(s) based on the selection rules described above, which may further include whether the text is associated with specialized field data (420) (e.g., accounting, legal, patent, tax, medical, financial, etc.).

[0183] In another embodiment, the language model (113) (or target conversation selection model) is trained to select a second target conversation, and the agent (112) can select the second target conversation using the language model (113) (or target conversation selection model). A training dataset that considers the characteristics of conversations using specialized data (420) can be used for training this language model (113) (or target conversation selection model). Similarly to the embodiments described above, for example, the Expert model (1132) can be trained to select a second target conversation from among conversations and generate output data for the corresponding first target conversation using specialized data (420). Similarly, training datasets that consider the characteristics of conversations using specialized data (420) can be additionally used for this training.

[0184] In step S1020, the agent (112) provides the second target conversation as input data to the language model (113), and in step S1030, the agent (112) can obtain output data for the second target conversation generated using the specialized data (420) from the language model (113). As in the embodiments described above, the Expert model (1132) is trained to respond to input using the specialized data (420), and thus, as the second target conversation is input from the agent (112), the Expert model (1132) can generate output data for the second target conversation using the specialized data (420).

[0185] In one embodiment, the agent (112) may provide input data including a second target conversation and conversations to the Expert model (1132), thereby requesting the language model (113) to generate output data for the first target conversation using the conversations. For example, the agent (112) may generate a prompt to instruct the Expert model (1132) to generate output data using conversations of meeting participants in a chat room, thereby obtaining output data generated using the details contained in the conversations and the domain data (420) from the Expert model (1132).

[0186] According to the above-described embodiment, the agent (112) corresponding to the AI ​​conference participant can provide specialized responses to the second target conversation using specialized field data (420) as an expert invited to the conference. Accordingly, based on learning about in-depth specialized knowledge and know-how focused on a specific field, the agent can function as an expert AI that can provide consulting for problem solving to the conference participants. Even without a separate expert taking the time to attend the conference, it can solve problems in specialized fields that are difficult for humans but possible for AI.

[0187] Referring again to FIG. 8, the Participant model (1133) according to one embodiment may be trained to engage in conversations with multiple users using public data. For example, the Participant model (1133) may be a generative AI model trained to dynamically participate in conversations between multiple users, and may be a conversational model implemented to enable context-sensitive conversations between multiple conference participants, whether in real-time or non-real-time.

[0188] In one embodiment, the Participant model (1133) may be trained using training data containing predefined reference conversations. In one embodiment, the Participant model (1133) may be obtained by applying additional training using the reference conversations to a generative AI model pre-trained to engage in conversations with users using public data.

[0189] Here, reference conversations represent learning conference conversations that include conversation sets by three or more users, and may include, for example, sets of diverse conversations that can occur in different organizational situations, similar to multi-party conversational materials that can occur in a conference. In one embodiment, the reference conversations may further include information about at least one of the meeting context (e.g., ideation, project phase establishment, etc.), the relevant organization (e.g., company, school, military, private organization, etc.), and the user context (e.g., user-specific roles, positions, work experience, etc.) in which the conversations take place, along with the conversations by three or more users. For example, in a project planning meeting at a company related to a certain industry, a situation in which related conversations are exchanged between three participants of the same or different positions may be assumed, and the reference conversations by these participants may be prepared from conversation samples acquired according to a predefined method (e.g., designer's design, collection of public data, etc.). For another example, in a meeting for ideation on a given meeting topic at an educational institution, reference conversations can be prepared by assuming a situation in which related conversations are exchanged among five participants of the same or different ranks.

[0190] The Participant model (1133) according to one embodiment may be trained by a supervised learning, semi-supervised learning, or self-supervised learning method using training data including the above-described plurality of reference conversations. For example, during the training process of the Participant model (1133), training may be performed in a self-supervision manner in which, without labeling the correct answer (i.e., training correct answer data), a random participant is selected from among the reference conversations as training input data, the conversation of the participant is masked, and a task is assigned to match the masked conversation by considering the above-described organization and situation, etc. In another example, each reference conversation may include conversations corresponding to training input data and conversation responses corresponding to training correct answer data, and training may be performed in a supervised learning method in which each reference conversation is used as training data during the training process of the Participant model (1133).

[0191] The Participant model (1133) according to the aforementioned embodiments can learn conference conversation skills that enable it to independently determine and respond to conversations requiring a response by comprehensively considering the content of the conversations among the meeting participants, the relevant organization, and the context, when given conversations. Accordingly, when conversations are input, the Participant model (1133) can generate output data using public data for conversations requiring a response.

[0192] FIG. 11 illustrates an exemplary flowchart for an agent (112) according to a third embodiment to obtain output data for a target conversation using public data (430).

[0193] Referring to FIG. 11, in step S1110, the agent (112) can select a third target conversation that includes a conversation requiring a response using public data (430) from among the acquired conversation(s).

[0194] In one embodiment, the agent (112) may select the third target conversation based on the selection rules described above. For example, the agent (112) may select the remaining conversations that do not fall into the first or second target conversations as the third target conversations. In another embodiment, the agent (112) may use the Participant model (1133) to select the third target conversations, and may, for example, pass the acquired conversations to the Participant model (1133).

[0195] In step S1120, the agent (112) provides the third target conversation as input data to the language model (113), and in step S1130, the agent (112) can obtain output data for the third target conversation generated using the public data (430) from the language model (113). In one embodiment, step S1120 can be performed together with step S1110, and for example, the agent (112) can provide the obtained conversations to the Participant model (1133) and request the Participant model (1133) to generate output data for conversations requiring a response among the conversations.

[0196] According to the above-described embodiment, the agent (112) can function as a complete participant that autonomously participates in a conversation in which multiple people participate according to the situation and organization without interrupting the conversation flow by utilizing the Participant model (1133) that has learned conversation skills for multi-party meetings.

[0197] According to one embodiment, a third target conversation may be a conversation requiring a response using recently disclosed data. Here, recently disclosed data may refer to recent disclosed data not included in the public data (430).

[0198] In one embodiment, the agent (112) may select a third target conversation that requires a response using recent public data from among the conversation(s) based on selection rules that further include whether the conversation(s) contain text associated with recent public data (e.g., recent trends, updated material, current affairs, etc.).

[0199] In one embodiment, if a third target conversation is selected as a conversation requiring a response using recent public data, the agent (112) may obtain recent public data associated with the third target conversation and cause the Participant model (1133) to generate output data using the obtained recent public data. For example, the recent public data included in the database (400) may be updated to include recent data according to a preset cycle, and the agent (112) may use a keyword search method to obtain search results for recent public data associated with the third target conversation from the database (400). In another example, the agent (112) may use a preset data collection method (e.g., communication with specific data servers for data collection, data collection through crawling, etc.) to obtain search results for recent public data associated with the third target conversation from external devices. The agent (112) may generate a prompt to instruct the Participant model (1133) to generate output data using the obtained search results. Alternatively, the agent (112) may use the acquired search results to enable additional learning of the Participant model (1133), and provide a third target conversation to the additionally learned language model (113), thereby obtaining output data using recently public data.

[0200] Meanwhile, at least some of the above-described embodiments may be implemented in a mixed form, and are not limited to the examples described above. For example, the agent (112) may select a target conversation requiring a response using at least some of the local internal data (410), specialized data (420), and public data (430) from among the acquired conversation(s), and request the Expert model (1132) to generate a response to the target conversation using the search results for the local internal data (410) associated with the target conversation. As another example, the language model (113) may include at least some of the Colleague model (1131), the Expert model (1132), and the Participant model (1133), or may be a mixed model obtained by applying at least some of the above-described learning methods.

[0201] According to one embodiment of the present disclosure, the language model (113) can generate output data in at least one response form selected from the group consisting of questions, answers, opinions, and summaries. The response form may be determined by the agent (112) or by the language model (113).

[0202] In one embodiment, the response form may be determined by the agent (112). Specifically, the agent (112) may select a target conversation and determine the response form of the target conversation based on the target conversation. In addition, the agent (112) may provide the target conversation and the determined response form together to the language model (113), and request the language model (113) to generate output data for the target conversation corresponding to the response form. In one embodiment, the agent (112) may determine the response form based at least in part on the selection rule described above. For example, the agent (112) may determine the response form based on at least one of the type of conversation, the intent of the conversation, the context of the preceding conversations, the time interval between the conversations, and the number of conversations during a predetermined time period, as applied to the selection rule described above. For example, the agent (112) may determine a response form (e.g., an answer to a question, an opinion to a request, etc.) corresponding to the type of the target conversation (e.g., a question, a request, etc.), determine a response form (e.g., sharing the result of an action, sharing information) corresponding to the intention of the target conversation (e.g., suggesting an action, confirming information, etc.), determine a response form considering the context of previous conversations (e.g., presenting an opinion in the case of a conversation flow of sharing opinions), determine a response form considering the time interval or number of conversations (e.g., sharing (opinion) of recent information related to the meeting agenda in the case of a conversation flow of sharing opinions or sharing ...

[0203] In another embodiment, the response format may be determined by the language model (113) (or target conversation selection model). For example, the Participant model (1133) may be trained to generate output data in the form of a response of one of questions, answers, opinions, and summaries for conversations requiring a response among the conversations. Each of the reference conversations used to train the Participant model (1133) includes conversations in two or more of the following types: questions, answers, opinions, and summaries, and the Participant model (1133) may be trained to generate output in the form of a response of questions, answers, opinions, or summaries appropriate to the organization and situation from each of the reference conversations. The agent (112) transmits conversations or target conversations to the language model (113), and the language model (113) may determine which of the transmitted conversations requires a response and generate a response in the appropriate response format. As another example, the target conversation selection model may be a model trained to determine both the target conversation and the response format of the target conversation from the conversations.

[0204] Meanwhile, an agent (112) according to one embodiment may generate output data corresponding to the setting information of an AI conference participant as a response to a target conversation. In one embodiment, the conference room provider (111) may, in the process of receiving an invitation request for an AI conference participant from a host terminal (200) or a participant terminal (300), also receive the setting information of the AI ​​conference participant, and may set an agent (112) participating in the online conference room (10) to participate in the conference based on the setting information of the AI ​​conference participant.

[0205] In one embodiment, the configuration information of the AI ​​meeting participant includes configuration information about the decision method of the response form and / or access rights, and may further include configuration information about at least one of name, title, affiliation, tone of voice, age, gender, and language.

[0206] Here, the method for determining the response type refers to a method for assigning different weights to at least one of a question, an answer, an opinion, and a summary in the process of determining the response type of output data for the target conversation. For example, the method for determining the response type may include at least one of a question participation type in which questions are given higher weights than other types, an answer participation type in which answers are given higher weights, an opinion participation type in which opinions are given higher weights, and a summary participation type in which summary is given higher weights. For example, when selecting a target conversation, the agent (112) may select a conversation whose response type of output data corresponds to the corresponding setting information (e.g., answer-centered) among the conversations (e.g., a conversation whose conversation type is question) with a relatively higher probability than other conversations, and when determining the response type of output data, the response type (e.g., answer) corresponding to the corresponding setting information (e.g., answer-centered) may be selected with a higher probability than other types. As another example, when considering the decision method of the response form (e.g., answer-centered) among conversations, the language model (113) can be trained to generate output data in the response form (e.g., answer) corresponding to the decision method of the response form with a relatively higher probability for conversations requiring a response (e.g., question).

[0207] In addition, the access right indicates the access right granted to the agent (112) corresponding to the AI ​​conference participant, and can be used to restrict the use of the local internal data (410) to which the security level corresponding to the access right is granted when a response to the target conversation is generated using the local internal data (410). In one embodiment, when the conference host and conference participants participating in the online conference room (10) are determined, the access right of the AI ​​conference participant participating in the online conference room (10) can be determined based on the access right of each of the conference host and conference participants according to the pre-stored security policy.

[0208] In one embodiment, the agent (112) may provide the AI ​​conference participant's configuration information along with the target conversation to the language model (113), thereby requesting the language model (113) to generate a response to the target conversation by considering the configuration information. In another embodiment, individually trained language models (113) may be prepared to generate responses corresponding to each of a plurality of configuration cases that can be combined in the AI ​​conference participant's configuration information, and the agent (112) may communicate with the language model (113) corresponding to the configuration information to obtain a response that matches the configuration information.

[0209] In one embodiment, the agent (112) may acquire a profile image corresponding to the AI ​​conference participant's configuration information from among a plurality of pre-stored profile images, and may provide a conversation response to the target conversation along with the corresponding profile image. In one embodiment, the remaining configuration information of the AI ​​conference participant, excluding access rights, may be updated based on an update request from the host terminal (200) or the participant terminal (300).

[0210] According to one embodiment of the present disclosure, the agent (112) may provide a conversation response for a target conversation to a participant terminal (300) of a conference participant based on the acquired output data. In one embodiment, the agent (112) may generate a conversation response including the output data or a conversation response processed, converted, or modified from the output data, and may provide the generated conversation response to the participant terminal (300) accessing the online conference room (10). This will be further described with reference to FIG. 12.

[0211] Fig. 12 is a diagram illustrating an example of a conversation response (30) displayed on a screen of a participant terminal (300) according to one embodiment. Fig. 12 illustrates a case in which an agent's (112) conversation response (30) is provided in the form of a chat message through a messenger-based chat user interface (UI). However, this is merely an example, and depending on the implementation, the conversation response (30) may be implemented in various ways, such as being provided in the form of a comment through a comment-based conversation UI (e.g., comment-based YouTube) or being provided in the form of a data file.

[0212] Referring to FIG. 12, the agent (112) may provide a conversation response (30) to each participant terminal (300) of a conference participant or each participant terminal (300) of the conference participants in the form of a message by an AI conference participant. For example, the agent (112) may add the conversation response (30) to the conversation thread object (13) of the online conference room (10) in the form of a message by an AI conference participant based on the granted conversation input authority. Accordingly, the conversation response (30) may be displayed as a conversation by an AI conference participant participating in the chat room on each of the participant terminals (300) accessing the online conference room (10).

[0213] In one embodiment, the agent (112) may provide a conversation response (30) to the participant terminal (300) of the conference participant or each of the participant terminals (300) of the conference participants in the form of a message by an AI conference participant that is visually distinct from the conference participants. For example, the display method of the conversation response (30), such as the color of the message, the speech bubble shape, the font shape, or the font size, may be set to be at least partially different from the display method of the conference participants. As another example, a participant profile image representing the AI ​​conference participant may be provided along with the conversation response (30), and the participant profile image or its display method may be set to be visually distinct from the conference participants. Accordingly, the conference participants can easily visually check the message of the AI ​​conference participant in the chat room.

[0214] In one embodiment, the agent (112) may provide information about a target conversation to the participant terminal (300) along with a conversation response (30). For example, if a second message (22) is selected as a target conversation among the conversations (20) and a conversation response (30) for the second message (22) is generated, the agent (112) may cause information indicating that the target conversation of the conversation response (30) is the second message (22) (e.g., indicating that it is a reply message for the second message (22)) to be displayed along with the conversation response (30).

[0215] In one embodiment, the agent (112) may provide the response form of the output data to the participant terminal (300) together with the conversation response (30). For example, the agent (112) may cause display information (e.g., text, image, icon) representing any one of the response forms of the conversation response (30) among a question, an answer, an opinion, and a summary to be displayed together with the conversation response (30).

[0216] In one embodiment, the agent (112) may provide the participant terminal (300) with a response basis (40) including the basis data used to obtain the output data and / or an access link to the basis data, together with the conversation response (30). Specifically, the agent (112) may provide the response basis (40) with the basis data and / or the access link based on the access rights granted to each conference participant. For example, when generating a prompt, the agent (112) may generate a prompt that further includes content instructing to additionally provide information about the basis data used to obtain the output data and / or the corresponding access link. In addition, the agent (112) may extract the basis data and / or the access link from the output data, and provide the conversation response (30) included in the output data and the response basis (40) including the extracted basis data and / or the access link in the form of a message.

[0217] In one embodiment, the agent (112) may determine whether to provide a conversation response (30) based on the access rights granted to each conference participant. For example, the agent (112) may perform a double check to confirm whether the security level corresponding to the basis data provided from the language model (113) corresponds to the access rights granted to each conference participant (or the access rights granted to the online conference room (10) from this), and if they do not correspond, the conversation response (30) may not be provided to the participant terminals (300) of the conference participants. In another example, the conversation response (30) may be provided only to the participant terminals (300) of the conference participants who have access rights corresponding to the security level corresponding to the basis data among the conference participants.

[0218] In one embodiment, the agent (112) may obtain a plurality of responses corresponding to the security levels corresponding to the access rights of the meeting participants for the target conversation from the language model (113), and provide each response corresponding to the access rights of each meeting participant to each participant terminal (300). Specifically, when the security levels corresponding to the access rights granted to the meeting participants are different, the agent (112) may obtain a plurality of responses corresponding to the security levels for the target conversation from the language model (113). For example, if a first meeting participant of a higher position and a second meeting participant of a lower position (e.g., manager) participate in a meeting, the agent (112) may obtain a first search result for local internal data (410) with a first security level corresponding to the higher position and a second search result for local internal data (410) with a second security level corresponding to the lower position, and may cause the language model (113) to generate first output data (e.g., an answer generated based on corporate data with a relatively high security level) and second output data (e.g., an answer generated based on corporate data with a relatively low security level) using the first search result and the second search result, respectively. The agent (112) may process each output data as a conversation response accessible to a meeting participant with access rights corresponding to each security level and add it to the conversation thread object (13). Accordingly, on each participant terminal (300), the same or different answers may be displayed for the same question, or some answers may not be displayed, depending on the participation rights of each meeting participant.

[0219] In one embodiment, if the target conversation is a conversation for which a response has been individually requested by a participant terminal (300) of a specific conference participant, the agent (112) may provide the conversation response (30) only to the participant terminal (300) of the specific conference participant. For example, a conversation may be received with an individual response request for an AI conference participant through a general conference chat room in which conference participants participate, and the agent (112) may select the conversation as a target conversation and then securely process the conversation response (30) for the conversation so that only the specific conference participant can view it, and add it to the conversation thread of the general conference chat room included in the conversation thread object (13). As another example, in connection with an online conference room (10), a one-on-one conference chat room is created at the request of each conference participant, in which only each conference participant and the AI ​​conference participant participate, and a conversation can be received from the conference participant through the one-on-one conference chat room, and the agent (112) can regard the conversation as a target conversation and add a conversation response (30) to the conversation to the conversation thread of the entire one-on-one conference chat room included in the conversation thread object (13).

[0220] In one embodiment, the agent (112) may provide a conversation response (30) according to the input method of the conversation selected as the target conversation, or may provide the conversation response (30) in at least one preset output method among text, image, video, and audio. For example, if a conversation received in the audio input method is converted into text and provided to the chat room as a second message (22), and the second message (22) is selected as the target conversation, the conversation response (30) to the second message (22) may be converted into audio output and provided to the chat room in the form of a voice message by AI, or may be provided to the chat room in the form of a text message according to the preset output method.

[0221] According to one embodiment of the present disclosure, the additional learning unit (114) may be implemented to perform additional learning using meeting information and / or conversations in the online conference room (10). The additional learning according to one embodiment may include pre-learning performed before the start of the meeting and / or real-time learning performed during the meeting. The additional learning according to one embodiment may be understood as a concept encompassing additional training of a previously learned language model (113) post-learning to provide an optimized conversational response for the online conference room (10) of the agent (112). Embodiments related to this will be further described with reference to FIG. 13.

[0222] Figure 13 illustrates an exemplary flowchart in which pre-learning is performed by an additional learning unit (114) according to one embodiment.

[0223] Referring to FIG. 13, the additional learning unit (114) may perform pre-learning using meeting information of the online conference room (10) before the online conference room (10) is initiated. Here, the initiation of the online conference room (10) may mean the arrival of a start time during the conference period of the online conference room (10) or the receipt of a meeting start request. In one embodiment, the additional learning unit (114) may perform the above-described pre-learning when the participation of the agent (112) in the online conference room (10) is determined.

[0224] Specifically, in step S1310, the meeting room provider (111) may determine whether or not the agent (112) participates in the online meeting room (10). For example, as described above, if the meeting opening request includes an invitation request for an AI meeting participant, the meeting participant's authority for the online meeting room (10) may be granted to the agent (112) by the meeting room provider (111) before the meeting starts.

[0225] In step S1320, the additional learning unit (114) can obtain meeting information of the online conference room (10) after the participation of the agent (112) is determined. For example, the additional learning unit (114) can obtain meeting information from the basic information object (11) and the meeting history object (14) of the online conference room (10). In one embodiment, the meeting information of the online conference room (10) used for pre-learning may include at least one selected from the group consisting of (i) meeting period, (ii) meeting content, (iii) information on meeting participants, (iv) meeting classification information, and (v) a meeting history indicating a historical relationship between one or more other online conference rooms (10) associated with the online conference room (10). For example, the meeting information of the online conference room (10) includes the meeting period, meeting content, information on the meeting host, information on the meeting participants, meeting classification information, meeting title, and meeting objective, etc. included in the basic information object (11), and may further include the meeting history object (14).

[0226] In step S1330, the additional learning unit (114) may acquire external data associated with the meeting information of the online conference room (10) before the online conference room (10) is started. For example, the additional learning unit (114) may acquire external data associated with the meeting information of the online conference room (10) from one or more external devices using the above-described data collection method (e.g., communication with specific data servers for data collection, data collection through crawling, etc.). The acquired external data may include public data associated with the meeting information (e.g., technical field details), recent public data (e.g., recent industry trends), and / or specialized field data (420) (e.g., revisions to laws, etc.).

[0227] In step S1340, the additional learning unit (114) can perform pre-learning using the acquired meeting information of the online conference room (10) and / or acquired external data.

[0228] In one embodiment, the additional learning unit (114) may perform additional learning on the language model (113) using the acquired external data. For example, the additional learning unit (114) may preprocess the acquired external data as learning data by applying a predefined method, and may pre-train the language model (113) using the pre-processed learning data. For example, the additional learning unit (114) may collect expert knowledge on recent industry trends or related technologies related to the meeting topic before the start of the meeting, and may perform additional learning on the language model (113) through observation and pattern matching processes on the learning data based on the collected expert knowledge. In one embodiment in which additional learning on the language model (113) is easy, by pre-training the language model (113) with additional information related to the meeting collected through various channels before the start of the meeting in this way, a language model (113) that better understands information related to the meeting can be prepared before the start of the meeting.

[0229] In another embodiment, the additional learning unit (114) can update the database (400) to include the acquired external data. For example, the additional learning unit (114) can process data transformation and indexing of external data associated with the corresponding meeting information to create a database so that it corresponds to the keyword search method and / or semantic search method used to obtain search results by the agent (112). Accordingly, in a subsequent process, the agent (112) can obtain search results from the database (400) that further includes external data associated with the meeting information, and the language model (113) can generate output data for the target conversation based on these search results. In an embodiment where additional learning of the language model (113) is difficult, an environment can be supported in which the language model (113) can provide an optimized response for the meeting by considering the relevant additional information in this manner.

[0230] In another embodiment, the additional learning unit (114) may store the acquired external data in the corresponding online conference room (10), and may additionally utilize the stored external data during the process of providing a conversation response by the agent (112). For example, the online conference room (10) may further include an additional information object (not shown) in which information for pre-learning and / or real-time learning is stored, and the additional learning unit (114) may include the acquired external data in the additional information object of the online conference room (10). In the subsequent process, the agent (112) may additionally utilize the external data associated with the meeting information included in the additional information object to generate a prompt requesting the generation of output data for the target conversation, and provide the prompt to the language model (113). In this case, the information stored in the additional information object is not learned by the language model (113), but is utilized only in the corresponding online conference room (10), and may be deleted when the meeting ends, which is advantageous in terms of information security.

[0231] In another embodiment, the additional learning unit (114) can perform additional learning on the language model (113) using the acquired meeting information. For example, the additional learning unit (114) can acquire the historical relationships between the current meeting and other meetings, as well as meeting information on other meetings, based on the meeting history object (14), and can use the acquired information to pre-train the language model (113) to understand the meeting history between past meetings, meeting result reports from past meetings, and work items for each participant, which will be described later. Accordingly, a language model (113) can be prepared that can remember and track the history and work items of past meetings related to the current meeting before the meeting begins.

[0232] In one embodiment, conversations from meetings prior to the creation of the online conference room (10) may be additionally utilized for the above-described prior learning. Specifically, the additional learning unit (114) may acquire conversations from other meetings associated with the online conference room (10) based on the meeting history object (14) and perform prior learning additionally using the conversations. For example, in the case of prior learning for the first online conference room, the additional learning unit (114) may search for other second online conference rooms associated with the first online conference room from the meeting history object (14) of the first online conference room, which is the current meeting, and may acquire conversations from meetings prior to the creation of the first online conference room from the conversation thread object (13) of each of the second online conference rooms, which are past meetings.

[0233] In one embodiment, the additional learning unit (114) may perform pre-training on the language model (113) using conversations from meetings prior to the creation of the acquired online conference room (10). Similarly, the additional learning unit (114) may preprocess the acquired past conversations as training data by applying a predefined method, and may perform additional training on the language model (113) using the pre-processed training data. Accordingly, a language model (113) that has a good understanding of the level of meeting goal achievement, previously derived outcomes, related knowledge, and factual relationships contained in conversations from past meetings related to the meeting can be prepared before the start of the meeting.

[0234] In another embodiment, the additional learning unit (114) may include conversations from meetings prior to the creation of the acquired online conference room (10) in the local internal data (410). Accordingly, during the conference process, the agent (112) may obtain search results from the local internal data (410) that further include conversations from meetings prior to the creation of the corresponding online conference room (10), and provide conversation responses based on these search results.

[0235] In another embodiment, the additional learning unit (114) may store conversations from meetings prior to the creation of the acquired online conference room (10) in the online conference room (10), and may additionally use the conversations thereafter in the process of providing conversation responses by the agent (112). For example, as described above, the additional learning unit (114) may include conversations from meetings prior to the creation of the acquired online conference room (10) in the additional information object of the online conference room (10), and in the subsequent process, when the agent (112) transmits the target conversation to the language model (113), the agent (112) may also provide information included in the additional information object, and may request the language model (113) to additionally use the information to generate output data for the target conversation.

[0236] According to the embodiments described above, since prior learning is performed on external data related to the meeting information, meeting history, conversations from past meetings, etc. before the meeting starts, the agent (112) can provide more customized responses to conversations occurring in the meeting.

[0237] An additional learning unit (114) according to one embodiment can perform real-time learning using conversations of meeting participants while an online conference room (10) is provided. Here, real-time learning may mean learning using conversations acquired in real time through the online conference room (10) or learning using conversations acquired during a predetermined time period corresponding to real time.

[0238] In one embodiment, the additional learning unit (114) can perform real-time learning on the language model (113) using conversations acquired through the online conference room (10). For example, the additional learning unit (114) can preprocess conversations added to the conversation thread object (13) as learning data by applying a predefined method, and can use the preprocessed learning data to train the language model (113) in real-time. The above-described real-time learning can be performed before, together with, or after providing a conversation response to the target conversation. Accordingly, the language model (113) can be trained to provide an appropriate response by comprehensively considering relevant knowledge and factual relationships included in the real-time conversations.

[0239] In another embodiment, real-time learning using conversations can be performed by the agent (112). For example, when providing a target conversation to the language model (113), the agent (112) can also provide conversations, requesting the language model (113) to generate output data based on additional learning using the conversations. In this case, additional learning is performed in the language model (113), and the language model (113) can understand and respond to facts, progress, etc. from the conversations exchanged in the meeting. In another example, the agent (112) can request the language model (113) to generate output data for the target conversation by additionally utilizing the conversations and / or meeting information from the online conference room (10). In this case, even if additional learning is not performed in the language model (113), the language model (113) can respond by considering the meeting atmosphere, special terms, conversation situations, etc. from the conversations in each meeting.

[0240] According to the embodiments described above, as real-time learning is performed on conversations added in real time and materials shared in the meeting while the meeting is in progress, the agent (112) can provide a more optimized response for the meeting.

[0241] Meanwhile, information used for pre-learning and / or real-time learning according to one embodiment may be stored and managed separately in the data space of the online conference room (10), and when the conference period of the online conference room (10) ends, use of the information may be restricted or prohibited, and for example, may be deleted when the conference end time arrives.

[0242] According to one embodiment of the present disclosure, the agent (112) can obtain a conference result report for an online conference room (10) based on conversations of conference participants. In one embodiment, the agent (112) can generate a conference result report summarizing the conference result in the online conference room (10) based on at least one of a basic information object (11), a content object (12), a conversation thread object (13), and a conference history object (14) included in the online conference room (10) when the conference corresponding to the online conference room (10) ends. The end of the conference according to one embodiment corresponds to the arrival of the end time during the conference period of the online conference room (10), or a conference end request from a host terminal (200) or a participant terminal (300).

[0243] In one embodiment, the agent (112) may obtain a meeting results report including summary results for each of a plurality of preset criteria items from the conversations. For example, the agent (112) may perform a summary for each criteria item for all conversations obtained through the online conference room (10), obtain a summary result, and apply the summary result to a preset report format to generate a meeting results report. Here, the plurality of criteria items may include at least one selected from the group consisting of a meeting overview item, a meeting content item, a main conversation collection item, a task item, and a next meeting item.

[0244] In one embodiment, the meeting overview item is an item representing basic information about the meeting, and may include at least one of the following: meeting duration (or meeting start time and meeting end time), meeting space (e.g., connection information for an online conference room (10)), information about meeting participants (e.g., name, position, affiliation, etc.), and meeting objective. For example, the meeting overview item may be obtained from meeting information (e.g., meeting duration, meeting participants, meeting objective, etc.) included in the basic information object (110), and if the meeting information does not include information related to a specific item, the meeting overview item may be obtained by extracting conversation content corresponding to the item from conversations included in the conversation thread object (13).

[0245] In one embodiment, a meeting content item represents summary information regarding the content discussed at the meeting, and may include key topic items related to the meeting objectives and additional topic items unrelated to the meeting objectives. Each topic item may include at least one subitem of keywords, a topic description, a summary of key conversations (e.g., comments, questions, and answers), and a conclusion.

[0246] In one embodiment, a "top conversation" item represents a sorting result regarding top conversations. A top conversation may be associated with a relatively large number of conversations (e.g., a source of replies, re-replies, etc.) or a conversation in which a relatively large number of meeting participants have selected a response (e.g., a "like" or "recommendation"). In one embodiment, a top conversation may be a conversation related to a top topic among the conversations, or a conversation regarding an additional topic that accounts for a predetermined percentage of the total number of conversations.

[0247] In one embodiment, the main conversation collection item may include items of conversation collections by conversation type (e.g., by question, by opinion, by answer) and / or related conversation collections (e.g., a collection of a specific question and a series of opinions, answers, etc. related to it). For the former example, conversations in a chat room may be classified into any one of the conversation types among question, opinion, answer, and others, and the agent (112) may generate a result that arranges the conversations classified by each conversation type in a tabular form as a summary result corresponding to the conversation collection item by conversation type. For the latter example, if a conversation in a chat room is related to other conversations, the interrelationships of those conversations may be managed (e.g., reply messages, re-ripple messages for a specific conversation, etc.), and the agent (112) may group the interrelated conversations as a summary result corresponding to the related conversation collection item by conversation type, and generate a result that arranges them for each group by chronological order, participants, or conversation type.

[0248] In one embodiment, a work item is an item representing summary information regarding a task assigned to one or more meeting participants among the matters discussed in the meeting, and may include at least one of information about one or more meeting participants, task content, task duration, task progress (e.g., progress rate, progress stage), and task results (e.g., description of results, task files, etc.). In one embodiment, each work item may be managed by follow-up actions of an agent (112), as will be described below.

[0249] In one embodiment, the next meeting item represents scheduled information regarding the next meeting, and may include at least one of the following: the next meeting schedule, the next meeting goal, and the next meeting participants. In one embodiment, if a summary result corresponding to at least one of the above-described items is not obtained from the conversations, the agent (112) may add a message inquiring about the item to the conversation thread object (13), and generate the summary result based on the response of the host terminal (200) or the participant terminal (300) to the message.

[0250] In one embodiment, the agent (112) may include one or more artificial intelligence models (e.g., a generative AI model optimized for summarization) trained to generate summary results for each of the aforementioned multiple criteria items from conversations. In another embodiment, the agent (112) may provide conversations to the language model (113), request the language model (113) to generate the aforementioned summary results from the conversations, and generate a meeting results report including the summary results obtained from the language model (113).

[0251] In one embodiment, the agent (112) may provide the meeting results report to the participant terminal (300) by adding it to the conversation thread object (13). Accordingly, meeting participants may view the meeting results report in the form of a message in the chat room. In another embodiment, the agent (112) may include the meeting results report in the basic information object (11). Accordingly, meeting participants may view the meeting results report included in the meeting information of the online conference room (10) in the form of a post.

[0252] In one embodiment, the agent (112) may modify the contents of the meeting result report based on a modification request from the host terminal (200) or participant terminal (300) for the meeting result report, and update the meeting result report so that the contents before and after modification and the modified meeting participant are clearly displayed.

[0253] Meanwhile, according to one embodiment of the present disclosure, the agent (112) may perform follow-up actions related to the online conference room (10) even after the end time of the conference period of the online conference room (10) arrives (or a request to end the conference is received). Follow-up actions according to one embodiment may include, but are not limited to, scheduling the next conference item and sharing the progress of a task item.

[0254] In one embodiment, the agent (112) may schedule the next meeting item based on the meeting participant's approval response to the next meeting item. For example, as the next meeting item in the meeting result report for the first online meeting room corresponding to the current meeting, the agent may create a second online meeting room that includes the next meeting schedule, the next meeting goal, and the summary results for each of the next meeting participants as meeting information, and configure the second online meeting room to be provided to each participant terminal (300) of the meeting participants according to the meeting schedule.

[0255] In one embodiment, the agent (112) may add a message to the conversation thread object (13) asking whether to perform scheduling based on the summary results corresponding to the next meeting item included in the meeting result report, and may perform the scheduling in response to an approval response from the host terminal (200) or participant terminal (300) to the message.

[0256] In one embodiment, the agent (112) may perform tracking management for each work item based on information about the work progress and / or work results received from the participant terminal (300) associated with each work item. The agent (112) may include information about each work item in the basic information object (11) based on the summary result corresponding to each work item, and perform tracking management for each work item. For example, the agent (112) may provide each participant terminal (300) with a message requesting an update of information about the work item during the work period of each work item, and may update the information about the work item in response to an update request about the work progress and / or work results received from the participant terminal (300). In one embodiment, the agent (112) may add information about the updated work item to the conversation thread object (13), thereby allowing the progress or results of each work item to be shared in the chat room.

[0257] In one embodiment, the agent (112) adds a message to the conversation thread object (13) asking whether to perform tracking management for a work item based on a summary result corresponding to the work item included in the meeting result report, and performs tracking management for the work item in response to an approval response from the host terminal (200) or the participant terminal (300) associated with each work item for the message.

[0258] When the above-described follow-up actions according to one embodiment are performed, for example, when the progress stage of the work items reaches completion, the session for conducting the meeting in the online conference room (10) may be terminated and the conference participant's authority granted to the agent (112) associated with the online conference room (10) may be released.

[0259] According to the above-described embodiments, meeting participants can conveniently review a meeting results report summarizing the conversations from the AI ​​participant conversations at the end of the meeting. Furthermore, they can easily check the progress of assigned work items even after the meeting concludes. In particular, in a meeting environment where multiple meetings are held with a specific goal, such as a task force, this system has the advantage of effectively tracking and managing the work items of participants generated from multiple meetings.

[0260] According to one embodiment of the present disclosure, the security unit (115) may be implemented to perform security procedures to enhance the security level according to pre-stored organization-specific security policies during provision of an online conference room (10). In one embodiment, the security unit (115) may provide a conversation response based on the results of performing additional security procedures for conference participants when local internal data (410) is used to provide a conversation response.

[0261] In one embodiment, the security unit (115) may request the provision of biometric information (e.g., voice, face, fingerprint, etc.) to each participant terminal (300) participating in the online conference room (10) and perform a security procedure to check whether the biometric information received from each participant terminal (300) corresponds to the biometric information previously registered in the user account of each participant. For example, the security procedure may be performed when a conference participant first accesses the online conference room (10), at preset time intervals, or each time the conference participant accesses the online conference room (10).

[0262] In another embodiment, the security unit (115) may perform a security procedure to determine whether at least one of the participant terminals (300) is based on an online conference environment combined with an offline conference. For example, if at least one participant terminal (300) is in an online conference environment combined with an offline conference, the security unit (115) may update the online conference room (10) to have participation rights corresponding to the lowest security level. In another embodiment, the security unit (115) may perform a security procedure to determine the identity and / or organization membership of each conference participant by verifying identity and / or authentication of a one-time password granted by an administrator.

[0263] In one embodiment, the security unit (115) can store and manage security policies for each organization. Security policies may include, for example, settings for security levels and participation permissions, and methods for implementing the aforementioned security procedures. Furthermore, the security unit (115) can monitor agents (112) and / or language models (113) based on the security policies. For example, the security unit (115) can store and manage the results of monitoring whether the access rights of meeting participants correspond with the security levels used to provide each conversation response.

[0264] As described above, the use of local internal data (410) requires a significant level of security. Therefore, in addition to the security procedures for restricting the use of local internal data (410) based on the aforementioned security level, an even higher level of security can be implemented through additional security procedures that verify the user's biometric information, meeting environment, etc. In some implementations, additional security procedures may also be implemented when specialized data (420) is used to provide conversational responses.

[0265] According to one embodiment of the present disclosure, the learning unit (116) may be implemented to perform learning on a language model (113). Learning by the learning unit (116) according to one embodiment may mean pre-training the language model (113) prior to the initiation of a conference service by the conference service device (100). Any portions overlapping with the above description will be omitted.

[0266] In one embodiment, the learning unit (116) can obtain a pre-trained model using a large amount of text data. For example, the learning unit (116) can perform pre-training on an initial language model through a task of predicting the next token based on cross entropy for a large amount of text data, or can receive a pre-trained model (e.g., a transformer-based language model) by receiving a model previously trained by an external device. In one embodiment, the learning unit (116) can obtain a pre-trained model to generate a natural language response to a natural language input, and for example, can obtain a pre-trained generative AI model to converse with a user in natural language.

[0267] In one embodiment, the learning unit (116) may obtain a plurality of learning datasets based on local internal data (410), and apply fine-tuning using the plurality of learning datasets to a pre-trained model to obtain a Colleague model (1131). For example, in the process of this fine-tuning, a plurality of first learning datasets may be used, and each first learning dataset may include learning input data including conversation(s) regarding the local internal data (410) and learning answer data including responses using the local internal data (410) associated with the conversation. In the learning according to one embodiment, various algorithms known as learning methods for language models or generative AI models, such as Large Language Models (LLM), Variational AutoEncoder (VAE), and Generative Adversarial Network (GAN), may be used.

[0268] In another embodiment, the learning unit (116) may obtain a plurality of learning datasets based on specialized field data (420) and apply fine-tuning using the plurality of learning datasets to a pre-trained model to obtain an Expert model (1132). For example, in the process of such fine-tuning, a plurality of second learning datasets may be used, and each second learning dataset may include learning input data including conversation(s) regarding a given specialized field and learning answer data including responses based on the specialized field data (420).

[0269] In another embodiment, the learning unit (116) may obtain a pre-trained model by using public data related to the field of expertise as learning data, and may apply fine-tuning using multiple learning datasets obtained from non-public data related to the field of expertise to the pre-trained model to obtain an Expert model (1132). For example, the learning unit (116) may obtain a pre-trained model by a self-supervised learning method using public data related to the accounting field (e.g., International Financial Reporting Standards, National Accounting Act, Local Accounting Act, etc.), and may implement the Expert model (1132) by fine-tuning the pre-trained model by a supervised learning method using multiple learning datasets preprocessed as learning data from professional knowledge, know-how, etc. provided by an accounting consulting company. In this way, rather than using a large amount of general public data, an Expert model (1132) that is well-versed in the relevant field and has expertise in problem solving can be implemented through intensive training using public and private data related to the relevant field.

[0270] In another embodiment, the learning unit (116) may obtain a plurality of learning datasets based on public data (430) and apply fine-tuning using the plurality of learning datasets to a pre-trained model to obtain an Expert model (1132). In one embodiment, the learning unit (116) may perform learning on a Participant model (1133) using learning data including predefined reference conversations. As described above, in the process of this fine-tuning, the predefined reference conversations are used as learning data, and additional learning using a supervised learning, semi-supervised learning, or self-supervised learning method may be performed to match the next conversation for each participant, thereby obtaining the Participant model (1133).

[0271] In another embodiment, the learning unit (116) can obtain a language model (113) as a result of performing learning by applying two or more learning methods among the Colleague model (1131), the Expert model (1132), and the Participant model (1133) described above.

[0272] Meanwhile, a process of creating an online conference room (10) based on a conference opening request according to one embodiment can be performed by a conference room provider (111).

[0273] Specifically, the conference room provider (111) may receive a conference opening request for an online conference room from the host terminal (200). In one embodiment, the conference opening request may include conference information including at least one of the conference period, conference content, conference participant information, and conference classification information.

[0274] The meeting duration includes information about the duration of a meeting, whether it is live or non-live. For example, the meeting duration for a live meeting may include information about the time. For another example, the meeting duration for a non-live meeting may include information about the start and end times, or information about the start time and time interval (e.g., 3 days, 7 days, 1 month).

[0275] The meeting content includes information about the topics to be discussed at the meeting. For example, the meeting content may include at least one of text data, image data, video data, and audio data to explain the meeting agenda, or an electronic document containing at least one of the data.

[0276] The meeting participant information includes information about users who can participate in the meeting, and may include, for example, user information (e.g., ID, name, company name, department name) of user accounts invited to the meeting. In one embodiment, the number of users corresponding to the meeting participant information may be set to be greater than or equal to N (where N is a natural number). For example, N may be greater than or equal to 3. In another example, N may be greater than or equal to 4 or greater than or equal to 2.

[0277] Meeting classification information includes one or more classification pieces of information associated with a meeting. For example, the meeting classification piece may include classification pieces indicating which technical field the meeting belongs to, or which task within a project the meeting is for. In one embodiment, the meeting classification piece may include classification pieces of information regarding at least one of the meeting organizer, meeting participant, department name, project name, a specific work or task within a workflow, a technical field, and past meetings prior to the meeting. As a specific example, the classification pieces of information regarding the meeting organizer, meeting participant, department name, project name, a specific work or task within a workflow, a technical field, and past meetings prior to the meeting may be defined through a preset classification system that distinguishes the scope to which the meeting belongs. In one embodiment, the classification system information may be read from memory (110) or generated or updated by an administrator account or a user account. In one embodiment, the classification scheme may include multiple levels having a hierarchical structure (e.g., where upper levels encompass lower levels) and / or a temporally sequential structure (e.g., where the temporal order of preceding and succeeding levels is distinguished).

[0278] In one embodiment, the above-described meeting information may further include, but is not limited to, a meeting title to implicitly indicate the meeting, a meeting objective to indicate the outcome sought or expected from the meeting, an introduction to the meeting host, or an introduction to the user group (e.g., company, department) to which the meeting host belongs.

[0279] The conference room provider (111) can create an online conference room (10) based on a received conference opening request. The creation of an online conference room (10) according to one embodiment can be understood as a concept corresponding to the opening of a conference. For example, the conference room provider (111) can create a conference identifier (e.g., conference ID) corresponding to the conference opening request, create an online conference room (10) corresponding to the conference identifier, and store and manage the same in a database (400).

[0280] The process of creating an online conference room (10) according to one embodiment may be performed in response to approval of a meeting opening request, but is not limited thereto. For example, the online conference room (10) may be created in response to approval of a meeting opening request. In another example, the online conference room (10) may be created as a meeting opening request or video data is acquired, and may be stored and managed in the database (400) in response to approval of the meeting opening request, or may be deleted in response to rejection of the meeting opening request. In addition, the database (400) may already store and manage information on a plurality of online conference rooms (10) corresponding to a plurality of meeting identifiers, as conferences created prior to the creation of the online conference room (10). These multiple online conference rooms (10) may be acquired and managed in the same or similar manner as described throughout the specification.

[0281] In the process of creating an online conference room (10) according to one embodiment, as described above, whether an agent (112) participates in the online conference room (10) can be determined. The conference room provider (111) can obtain invitation information for the agent's (112) participation in the conference during the process of receiving a conference opening request, and can allow the agent (112) to participate in the corresponding online conference room (10) based on the obtained invitation information. For example, an input menu for receiving conference information for a conference opening request and invitation information of an AI participant is displayed through a conference service application installed in the host terminal (200), and the conference room provider (111) can receive a conference opening request including the conference information entered by the host terminal (200) and the invitation information of the AI ​​participant through the input menu. The above-mentioned invitation information can include whether or not an AI conference participant is invited and / or setting information (e.g., a method for determining a response type, access rights, etc.).

[0282] The conference room provider (111) can confirm the participation response to the conference opening request of each conference participant based on the information of the conference participants. In one embodiment, the conference room provider (111) can transmit an invitation message including at least one of the conference period, conference content, conference host information, and conference participant information to each participant terminal (300) of the conference participants, and can confirm whether a participation response corresponding to the invitation message is received from each participant terminal (300).

[0283] If an invitation acceptance response corresponding to the invitation message is received from the participant terminal (300), the conference room provider (111) may determine that the participation response to the conference opening request of the corresponding conference participant has been confirmed. Similarly, if an invitation rejection response corresponding to the invitation message is received from the participant terminal (300), the conference room provider (111) may determine that the rejection response to the conference opening request of the corresponding conference participant has been confirmed.

[0284] The conference room provider (111) can determine whether to approve a conference opening request based on the participation responses to the conference opening request from each conference participant. In one embodiment, the conference room provider (111) can approve a conference opening request if a participation response corresponding to an invitation message is received from each participant terminal (300) of M (M is a natural number) (e.g., M is 2 or more) of the conference participants within a predetermined period of time.

[0285] The conference room provider (111) can acquire video data. According to one embodiment, the video data refers to data corresponding to a preset video format, and may include, for example, data corresponding to any known file format used to store digital video data in a typical computing environment. The conference room provider (111) can include the acquired video data in the content object (12) of the online conference room.

[0286] In one embodiment, the conference room provider (111) may obtain video data from the host terminal (200) or another user's terminal (not shown) designated by the host terminal (200). The other user's terminal refers to, for example, a user terminal corresponding to any one of the user accounts using the conference service to which the conference host has granted video upload authority for the conference. For example, the terminal may be a participant terminal (300) of any of the conference participants, or a user terminal unrelated to the conference participants.

[0287] In another embodiment, when a request for video generation is received from the host terminal (200) (or the terminal of the other user as described above), the conference room provider (111) may generate video data based on data received through linkage with the host terminal (200) (or the terminal of the other user as described above). For example, the host terminal (200) may transmit a request for video generation to the communication unit (120) through an installed conference service application, and the conference room provider (111) may link with the host terminal (200) through the conference service application in response to the request for video generation, and may control video recording to be performed at the host terminal (200).

[0288] In another embodiment, the conference room provider (111) may obtain video data from the conference opening request if the request includes video data. For example, if the conference opening request or the conference content includes video data, the conference room provider (111) may obtain the video data from it.

[0289] The process of acquiring video data according to one embodiment may be performed prior to receiving a meeting creation request, during the meeting creation request process, or after receiving the meeting creation request. Furthermore, in conjunction with this process, a meeting creation request approval process based on the provision of the aforementioned invitation message and participation response may be performed.

[0290] In one embodiment, the conference room provider (111) may provide an online conference room (10) to each participant terminal (300) of the conference participants whose participation response to the online conference room (10) has been confirmed among the conference participants. For example, the conference room provider (111) may provide an online conference room (10) to the participant terminals (300) in real time or non-real time according to the embodiment described above, only for the participant terminals (300) whose participation response has been confirmed by sending an invitation acceptance response corresponding to the invitation message described above.

[0291] In one embodiment, the conference room provider (111) can manage conversations received from each conference participant so that they correspond to any one of a plurality of preset message types. In one embodiment, the plurality of message types can include at least one selected from the group including questions, answers, opinions, departments, participants, and keywords. For example, a chat room implemented through an online conference room (10) is provided with a message input field for receiving messages and a selection button for selecting the message type of the corresponding message, and the conference room provider (111) can receive each message and message type together, and store and manage each message in a corresponding conversation thread object (13) by associating each message with the corresponding message type. Accordingly, each message in the conversation thread object (13) of each of the plurality of online conference rooms (10) stored in the database (400) can be managed so as to correspond to any one of the plurality of message types.

[0292]

[0293] FIG. 14 illustrates an exemplary flowchart of a computer-implemented method performed by a conference service device (100) according to one embodiment, for providing an artificial intelligence-based conference service. FIG. 14 can be understood with reference to all of the embodiments described above.

[0294] In one embodiment, the steps illustrated in FIG. 14 may be performed by the conference service device (100). In another embodiment, the steps illustrated in FIG. 14 may be implemented by a single entity, such as in a server. In another embodiment, the steps illustrated in FIG. 14 may be implemented by multiple entities, such as in a first server where some of the steps illustrated in FIG. 14 are performed and in a second server (or terminal).

[0295] Referring to FIG. 14, in step S1410, an agent (112) may acquire conversations received from conference participants participating in an online conference room (10) including an agent (112) connected to a language model (113). In one embodiment, the agent (112) is a conference participant object participating in the online conference room (10) together with the conference participants, and may be granted the same authority to input and receive conversations as the conference participants.

[0296] In step S1420, the agent (112) can select a target conversation that requires a response using local internal data (410) from among the acquired conversations.

[0297] In one embodiment, the agent (112) may use selection rules to select target conversations that require a response using local internal data (410).

[0298] In another embodiment, the agent (112) may pass the acquired conversations to a language model (113), which may then select target conversations. The language model (113) may be trained to select conversations requiring a response using local internal data (410) based on multiple conversations and the need for a response.

[0299] An agent (112) according to one embodiment may select one of the following among the conversations: a first target conversation requiring a response using local internal data (410), a second target conversation requiring a response using specialized data (420), and a third target conversation requiring a response using public data (430).

[0300] In step S1430, the agent (112) provides the target conversation as input data to the language model (113), and in step S1440, the agent (112) can obtain output data from the language model (113). The language model (113) may generate a natural language response to the natural language input using local internal data (410). In one embodiment, the language model (113) may generate a natural language response to the natural language input using local internal data (410) whose security level at least partially corresponds to the access rights granted to each meeting participant. In one embodiment, the language model (113) may be a generative AI model.

[0301] In one embodiment, the language model (113) may be a pre-trained model using local internal data (410). In one embodiment, the language model (113) may be further trained (e.g., pre-trained) using meeting information and / or conversations of the online meeting room (10). The agent (112) may provide a target conversation to the language model (113) trained using local internal data (410), and the language model (113) may generate output data for the target conversation. In one embodiment, the meeting information of the online meeting room (10) used for further training may include at least one selected from the group consisting of (i) the meeting duration, (ii) the meeting content, (iii) information on meeting participants, (iv) meeting classification information, and (v) a meeting history indicating a historical relationship between the online meeting room (10) and one or more other online meeting rooms (10).

[0302] In another embodiment, the agent (112) may obtain search results for local internal data (410) associated with a target conversation from the local internal data (410), generate a prompt to instruct the language model (113) to generate output data for the target conversation using the search results, and provide the generated prompt to the language model (113), thereby obtaining the output data from the language model (113). In one embodiment, the prompt may be for instructing the language model (113) to additionally generate the output data using conversations and / or meeting information from an online conference room.

[0303] In one embodiment, the language model (113) may be further trained (e.g., pre-trained) using conversations from meetings prior to the creation of the online meeting room (10). In another embodiment, the local internal data (410) may include conversations from meetings prior to the creation of the online meeting room (10).

[0304] In one embodiment, the agent (112) may provide target conversations and conversations to the language model (113), and request the language model (113) to generate a response based on additional learning (e.g., real-time learning) using the conversations. The language model (113) may generate output data for the target conversation based on additional learning (e.g., real-time learning) using the provided conversations.

[0305] In one embodiment, the language model (113) may be further trained using training data containing predefined reference conversations. In one embodiment, the reference conversations may include multiple conversations between multiple users across different organizational contexts. In one embodiment, the language model (113) may be obtained by applying additional training using the reference conversations as training data to a generative AI model pre-trained to engage in conversations with users using public data.

[0306] In one embodiment, the language model (113) may be a model trained using specialized data (410) associated with a preset specialized field. In one embodiment, the language model (113) may be obtained by applying additional training using specialized data (420) to a pre-trained generative AI model that is designed to converse with a user based on public data. In one embodiment, the language model (113) may be obtained by fine-tuning a model pre-trained using public data associated with the specialized field using private data associated with the specialized field.

[0307] In one embodiment, when a first target conversation is selected, the agent (112) provides the first target conversation as input data to a language model (113), for example, a Colleague model (1131), so as to obtain output data for the first target conversation generated using local internal data (410) from the Colleague model (1131). In one embodiment, when a second target conversation is selected, the agent (112) provides the second target conversation as input data to a language model (113), for example, an Expert model (1132), so as to obtain output data for the second target conversation generated using specialized data (420) from the Expert model (1132).

[0308] In one embodiment, the language model (113) may generate output data in at least one response form selected from the group consisting of questions, answers, comments, and summaries. In one embodiment, the response form may be determined by the language model (113) based on the target conversation, or by the agent (112) based on the target conversation.

[0309] In step S1450, the agent (112) may provide a conversation response to the target conversation to the participant terminal (300) of the conference participant based on the acquired output data. In one embodiment, the agent (112) may provide the conversation response to the participant terminal (300) in the form of a message by an AI conference participant that is visually distinct from the conference participants.

[0310] In one embodiment, the agent (112) may determine whether to provide a conversation response based on the access rights granted to each conference participant. In one embodiment, the agent (112) may provide, along with the conversation response, the basis data used to obtain the output data from among the local internal data (410) and / or an access link to the basis data. In one embodiment, if the target conversation is a conversation for which a response has been individually requested by a participant terminal (300) among the conference participants, the agent (112) may provide the conversation response only to the participant terminal (300).

[0311] In one embodiment, the agent (112) may obtain a meeting result report including summary results for each of a plurality of criteria items from the conversations, and provide the obtained meeting result report to the participant terminal (300). In one embodiment, the plurality of criteria items may include at least one selected from the group consisting of a meeting overview item, a meeting content item, a main conversation collection item, a task item, and a next meeting item.

[0312] In one embodiment, the agent (112) may perform scheduling for a meeting item based on a meeting participant's approval response for the next meeting item. And / or, in one embodiment, the agent (112) may perform tracking management for a work item based on information about the work progress and / or work results received from a participant terminal (300) associated with the work item.

[0313] According to one embodiment of the present disclosure, an AI conference participant is provided that provides reliable responses in a conference using local internal data (410). Furthermore, by selectively responding to conversations requiring a response, the AI ​​conference participant can seamlessly participate in multi-party conference conversations without disrupting the flow of conversation among conference participants. Consequently, the AI ​​conference participant can be provided as a colleague AI that is well-versed in local internal data (410) and capable of engaging in conversation, enabling equal participation in the conference with other conference participants.

[0314] According to another embodiment of the present disclosure, an AI conference participant can be provided that can provide expert consulting in a specific field using specialized field data (420). Based on intensive learning of specialized knowledge and know-how specific to a specific field, the AI ​​conference participant can be provided as an expert AI capable of providing consulting on various situations requiring problem solving in that field.

[0315] According to another embodiment of the present disclosure, an AI conference participant can be provided that utilizes organizational and context-specific reference conversations to engage in conversations appropriate to the organization and context in a meeting environment among multiple conference participants. Based on intensive learning of multi-party conference conversation skills, the AI ​​conference participant can be provided as a participatory AI that autonomously engages in conversations without interrupting the flow of conversation in meetings involving multiple participants, depending on the organization and context.

[0316] According to one embodiment of the present disclosure, a computer program stored in a computer-readable, non-transitory recording medium may be provided. The computer program includes instructions that, when executed by one or more processors (130), cause the one or more processors (130) to perform a computer-implemented method, the computer-implemented method comprising: a step in which the agent (112) acquires conversations received from conference participants participating in an online conference room (10) including an agent (112) connected to a language model (113); a step in which the agent (112) selects a target conversation requiring a response using local internal data (410) from among the acquired conversations; a step in which the agent (112) provides the target conversation as input data to the language model (113); the language model (113) generates a natural language response using the local internal data (410) for a natural language input; a step in which the agent (112) acquires output data from the language model (113); And the agent (112) may include a step of providing a conversation response to the target conversation to the participant terminal (300) based on the output data.

[0317] According to one embodiment of the present disclosure, a computer-readable, non-transitory recording medium storing a computer program may be provided. The computer program includes instructions that, when executed by one or more processors, cause the one or more processors to perform a computer-implemented method, the computer-implemented method comprising: a step in which the agent (112) acquires conversations received from conference participants participating in an online conference room (10) including an agent (112) connected to a language model (113); a step in which the agent (112) selects a target conversation requiring a response using local internal data (410) from among the acquired conversations; a step in which the agent (112) provides the target conversation as input data to the language model (113); the language model (113) generates a natural language response using the local internal data (410) in response to a natural language input; a step in which the agent (112) acquires output data from the language model (113); And the agent (112) may include a step of providing a conversation response to the target conversation to the participant terminal (300) based on the output data.

[0318] According to one embodiment of the present disclosure, a computer-implemented method may be provided that is performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor. The computer-implemented method comprises: a step of receiving a conference opening request including a conference period and information on conference participants from a host terminal (200); a step of creating an online conference room (10) including a conversation thread based on conversation input of conference participants based on the conference opening request; the online conference room (10) including an agent (112) connected to a language model (113), and the language model (113) generating a natural language response using local internal data (410) for a natural language input; a step of controlling the conference to proceed by non-real-time participation of conference participants by providing the online conference room (10) through a network in response to a request from each participant terminal (300) of the conference participants during the conference period; The method may include a step of having an agent (112) acquire conversations received from conference participants participating in an online conference room (10); a step of having the agent (112) select a target conversation requiring a response using local internal data (410) from among the acquired conversations; a step of having the agent (112) provide the target conversation as input data to a language model (113); a step of having the agent (112) acquire output data from the language model (113); and a step of having the agent (112) provide a conversation response for the target conversation to a participant terminal (300) based on the output data.

[0319] According to another embodiment of the present disclosure, a computer-implemented method may be provided that is performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor. The computer-implemented method comprises: a step in which the agent (112) acquires conversations received from conference participants participating in an online conference room (10) including an agent (112) connected to a language model (113); a step in which the agent (112) selects a target conversation requiring a response using specialized data (420) from among the acquired conversations; a step in which the agent (112) provides the target conversation as input data to the language model (113); the language model (113) generates a natural language response using the specialized data (420) for a natural language input; a step in which the agent (112) acquires output data from the language model (113); And the agent (112) may include a step of providing a conversation response for the target conversation to the participant terminal (300) of the conference participant based on the output data.

[0320] According to another embodiment of the present disclosure, a computer-implemented method may be provided that is performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor. The computer-implemented method comprises: a step in which the agent (112) acquires conversations received from conference participants participating in an online conference room (10) including an agent (112) connected to a language model (113); a step in which the agent (112) selects a target conversation requiring a response using public data (430) from among the acquired conversations; a step in which the agent (112) provides the target conversation as input data to the language model (113); the language model (113) is trained to generate a natural language response using the public data (430) for a natural language input; a step in which the agent (112) acquires output data from the language model (113); And the agent (112) may include a step of providing a conversation response for the target conversation to the participant terminal (300) of the conference participant based on the output data.

[0321] According to another embodiment of the present disclosure, a computer-implemented method may be provided that is performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor. The computer-implemented method may include a step in which the agent (112) acquires conversations received from conference participants participating in an online conference room (10) including an agent (112) connected to a language model (113); a step in which the agent (112) provides the acquired conversations as input data to the language model (113); the language model (113) is trained to generate natural language responses using training data including predefined reference conversations for natural language input; a step in which the agent (112) acquires output data for a conversation requiring a response among the conversations from the language model (113); and a step in which the agent (112) provides a conversation response to a participant terminal (300) of a conference participant based on the output data.

[0322] According to another embodiment of the present disclosure, a method for obtaining a language model (113) performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor may be provided. The method for obtaining the language model (113) may include: obtaining a pre-trained model to generate a natural language response to a natural language input; obtaining a plurality of training data sets based on local internal data (410); each training data set including training input data including a conversation about the local internal data (410) and training answer data including a response using the local internal data (410) associated with the conversation; and obtaining the language model (113) by applying fine-tuning using the plurality of training data sets to the pre-trained model.

[0323] According to another embodiment of the present disclosure, a method for obtaining a language model (113) performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor may be provided. The method for obtaining the language model (113) may include: obtaining a pre-trained model to generate a natural language response to a natural language input; obtaining a plurality of training data sets based on specialized data (420); each training data set including training input data including a conversation about a corresponding specialized field and training answer data including a response using specialized data (420) associated with the conversation; and obtaining the language model (113) by applying fine-tuning using the plurality of training data sets to the pre-trained model.

[0324] According to another embodiment of the present disclosure, a method for obtaining a language model (113) performed by a conference service device (100) including a memory (110), a processor (120), and one or more programs stored in the memory (130) and configured to be executed by the processor may be provided. The method for obtaining the language model (113) may include the steps of: obtaining a pre-trained model to generate a natural language response to a natural language input; obtaining training data including predefined reference conversations; and applying fine-tuning using the training data to the pre-trained model to obtain the language model (113).

[0325] As described above, the language model (113) can be trained to generate a natural language response using local internal data (410) for a natural language input using the method for acquiring the language model (113) described herein. This embodiment can be used independently of the technical feature for training the language model (113) without being combined with the technical feature for providing a conversational response using the pre-trained language model (113). Similarly, the technical feature for providing a conversational response using the language model (113) can be used independently of the technical feature for training the language model (113).

[0326] The steps illustrated above are merely examples, and the order, combination, branching, function, and performing entity may be added, omitted, or modified in various ways without departing from the essential characteristics of each technical feature described throughout the specification.

[0327] Meanwhile, the language model in this specification may refer to any form of computer program that operates based on a network function, an artificial neural network, and / or a neural network. Throughout this specification, the terms model, neural network, network function, and neural network may be used interchangeably. A neural network is a network in which one or more nodes are interconnected through one or more links to form input node and output node relationships within the neural network. The characteristics of the neural network can be determined based on the number of nodes and links within the neural network, the correlation between the nodes and links, and the weight value assigned to each link. A neural network may be composed of a set of one or more nodes. A subset of the nodes constituting the neural network may constitute a layer.

[0328] A deep neural network (DNN) may refer to a neural network that includes multiple hidden layers in addition to an input layer and an output layer. A deep neural network may include a convolutional neural network (CNN), a recurrent neural network (RNN), an autoencoder, a generative adversarial network (GAN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a Q network, a U network, a Siamese network, a generative adversarial network (GAN), a transformer, and the like. The description of the above-described deep neural network is merely an example, and the present disclosure is not limited thereto.

[0329] Neural networks can learn through at least one of the following methods: supervised learning, unsupervised learning, semi-supervised learning, self-supervised learning, or reinforcement learning. Neural network learning can be the process of applying knowledge to the neural network to perform a specific action.

[0330] Neural networks can be trained to minimize output errors. This process involves repeatedly inputting training data into the neural network, calculating the neural network output and target error for the training data, and backpropagating the neural network error from the output layer to the input layer to update the weights of each node in the neural network to reduce the error. In supervised learning, labeled data is used for each training data, while unsupervised learning uses unlabeled data. The amount of change in the connection weights of each updated node can be determined by the learning rate. The neural network's calculation of input data and backpropagation of errors can constitute a learning cycle (epoch). The learning rate can vary depending on the number of iterations in the neural network's training cycle. Additionally, to prevent overfitting, methods such as increasing the learning data, regularization, dropout that disables some nodes, and batch normalization layers can be applied.

[0331] In one embodiment, a language model may borrow at least a portion of a transformer. The transformer may be composed of an encoder that encodes embedded data and a decoder that decodes the encoded data. The transformer may have a structure that receives a series of data and outputs a series of data of different types through encoding and decoding steps. In one embodiment, the series of data may be processed into a form operable by the transformer. The process of processing the series of data into a form operable by the transformer may include an embedding process. Expressions such as data tokens, embedding vectors, and embedding tokens may refer to data embedded in a form operable by the transformer.

[0332] To encode and decode a series of data, a transformer can utilize an attention algorithm to process the encoders and decoders within the transformer. An attention algorithm can refer to an algorithm that, for a given query, calculates the similarity for one or more keys, reflects this similarity in the values ​​corresponding to each key, and then weights and sums the values ​​to which the similarity is reflected to calculate an attention value.

[0333] Depending on how the query, key, and value are configured, various types of attention algorithms can be categorized. For example, if attention is obtained by setting the query, key, and value all to the same value, this could be a self-attention algorithm. If attention is obtained by reducing the dimensionality of the embedding vector to process a series of input data in parallel and then generating individual attention heads for each segmented embedding vector, this could be a multi-head attention algorithm.

[0334] In one embodiment, the transformer may be composed of modules that perform multiple multi-head self-attention algorithms or multi-head encoder-decoder algorithms. In one embodiment, the transformer may also include additional components other than attention algorithms, such as embedding, normalization, and softmax. Methods for constructing a transformer using attention algorithms may include methods disclosed in Vaswani et al., Attention Is All You Need, 2017 NIPS, which is incorporated herein by reference.

[0335] A transformer can be applied to various data domains, such as embedded natural language, segmented image data, and audio waveforms, to transform a series of input data into a series of output data. To transform data with various data domains into a series of data that can be input to a transformer, the transformer can embed the data. The transformer can process additional data that expresses the relative positional relationship or phase relationship between the series of input data. Alternatively, vectors expressing the relative positional relationship or phase relationship between the input data can be additionally reflected in the series of input data to embed the series of input data. In one example, the relative positional relationship between the series of input data may include, but is not limited to, word order within a natural language sentence, the relative positional relationship between each segmented image, and the time order of segmented audio waveforms. The process of adding information expressing the relative positional relationship or phase relationship between the series of input data may be referred to as positional encoding.

[0336] In one embodiment, the language model may be a model trained using transfer learning. Transfer learning, in this context, refers to a learning method that pre-trains a large amount of unlabeled training data using semi-supervised or self-learning methods to obtain a pre-trained model for a first task, then fine-tunes the pre-trained model to suit a second task, and trains it on labeled training data using supervised learning to implement a target model.

[0337] In one embodiment, fine-tuning may encompass the concept of transferring a pre-trained model to a target task and subsequently training the model. Specifically, fine-tuning may refer to a method of modifying a pre-trained model to suit a task for generating conversational responses using a database (400) and updating the learning based on the weights of the pre-trained model. For example, such fine-tuning may include updating the parameters of a pre-trained model by additionally training the pre-trained model on a specific dataset for generating conversational responses.

[0338] Meanwhile, those skilled in the art will appreciate that the various exemplary logical blocks, modules, processors, means, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented by electronic hardware, various forms of programs or design code (referred to herein for convenience as software), or a combination of both. To clearly illustrate this interchangeability of hardware and software, various exemplary components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

[0339] The various embodiments presented herein can be implemented as a method, apparatus, or article of manufacture using standard programming and / or engineering techniques. The term article of manufacture includes a computer program, carrier, or media accessible from any computer-readable storage device. For example, computer-readable storage media include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic strips, etc.), optical disks (e.g., CDs, DVDs, etc.), smart cards, and flash memory devices (e.g., EEPROMs, cards, sticks, key drives, etc.). Furthermore, various storage media presented herein include one or more devices and / or other machine-readable media for storing information.

[0340] It should be understood that the specific order or hierarchy of steps in the presented processes is merely an example of exemplary approaches. It should be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure based on design priorities. The appended method claims provide elements of various steps in a sample order, but are not intended to be limited to the specific order or hierarchy presented.

Claims

1. A computer implementation method performed by a conference service device including a memory, a processor, and one or more programs stored in the memory and configured to be executed by the processor, A step in which the agent acquires conversations received from meeting participants participating in an online conference room including an agent connected to a language model; A step in which the agent selects a target conversation from among the conversations; the target conversation includes a conversation that requires a response using local internal data; A step in which the agent provides the target conversation as input data to the language model; wherein the language model generates a natural language response using the local internal data for the natural language input; A step in which the agent obtains output data for the input data from the language model; and The step of the agent providing a conversation response for the target conversation to the participant terminal of the conference participant based on the output data, Computer implementation method.

2. In paragraph 1, The above agent A conference participant object that participates in the online conference room together with the conference participants, characterized in that it is granted the same authority to input and receive conversations as the conference participants. Computer implementation method.

3. In paragraph 2, Further comprising a step of creating the online conference room based on a conference opening request from the host terminal, In the step of creating the above online conference room, characterized in that it is possible to determine whether the above agent participates, Computer implementation method.

4. In paragraph 3, After the participation of the agent is determined, the conference service device obtains conference information of the online conference room and obtains external data associated with the conference information before the online conference room is started; and Further comprising a step of performing additional learning on the language model using external data associated with the above meeting information. Computer implementation method.

5. In paragraph 1, The step of selecting the above target conversation is: The agent is characterized in that it selects the target conversation using a preset selection rule, and the selection rule is based on at least one selected from the group consisting of the type of conversation, the intention of the conversation, whether the other meeting participant is nominated, whether the AI ​​(Artificial Intelligence) meeting participant is nominated, the context of the conversations, the time interval of the conversations, the number of conversations during a predetermined time period, and whether the text associated with the local internal data is included. Computer implementation method.

6. In paragraph 5, In the step of providing the above target conversation to the language model, The agent provides the target conversation and the conversations to the language model, In the step where the above agent obtains output data from the above language model, The above language model generates the output data based on additional learning using the above conversations. Computer implementation method.

7. In paragraph 1, The above language model is trained to select conversations that require the above response based on multiple conversations and the necessity of the above response, The above agent transmits the received conversations to the language model, Computer implementation method.

8. In paragraph 1, The above local internal data is given a security level associated with the user's access rights, The above access rights are Characterized in that it is set by considering at least one selected from the group consisting of rank, position, document security level, department, employment type, workplace and affiliate. Computer implementation method.

9. In paragraph 8, The above language model is, A method of generating a natural language response to a natural language input by using local internal data of a security level corresponding to the access rights granted to each meeting participant among the above local internal data. Computer implementation method.

10. In paragraph 1, The above local internal data includes local internal data associated with an organization to which at least some of the meeting participants belong, Local internal data associated with the above organization is categorized based on one or more types selected from the group consisting of human resources, finance, accounting, legal, management, research, development, production and sales. Computer implementation method.

11. In paragraph 1, The above language model is characterized by being a generative artificial intelligence (AI) model. Computer implementation method.

12. In paragraph 11, The above generative AI model is, A pre-trained model using the above local internal data, Computer implementation method.

13. In paragraph 12, The above generative AI model is, Additional learning is done using the meeting information and / or conversations in the above online conference room. Computer implementation method.

14. In paragraph 13, The above local internal data is Including conversations from meetings prior to the creation of the above online conference room; Computer implementation method.

15. In paragraph 13, The above generative AI model is, Additional learning is made using conversations from meetings prior to the creation of the above online conference room. Computer implementation method.

16. In paragraph 13, The meeting information for the above online conference room is: At least one selected from the group consisting of (i) a meeting duration, (ii) meeting content, (iii) information on meeting participants, (iv) meeting classification information, and (v) a meeting history indicating a historical relationship between the online meeting room and one or more other online meeting rooms associated with the online meeting room. Computer implementation method.

17. In paragraph 11, The above generative AI model Additional training is performed using training data containing predefined reference conversations. Computer implementation method.

18. In paragraph 11, The above generative AI model is, A model trained using specialized data related to a preset specialized field. Computer implementation method.

19. In paragraph 11, The step of the above agent obtaining the above output data is: A step in which the agent obtains search results for local internal data associated with the target conversation from the local internal data; A step for generating a prompt for instructing the generative AI model to generate the output data for the target conversation using the search results; and A step in which the agent provides the prompt to the generative AI model, thereby obtaining the output data from the generative AI model. Computer implementation method.

20. In paragraph 19, The above prompt is, To instruct the generative AI model to additionally generate the output data by using the above conversations and / or meeting information of the online conference room. Computer implementation method.

21. In paragraph 1, The step of selecting the above target conversation is: The agent selects one of the first target conversation requiring a response using local internal data and the second target conversation requiring a response using specialized data among the above conversations, The above language model is, A first model that generates a natural language response using the local internal data for natural language input and a second model that is trained to generate a natural language response using specialized field data associated with a preset specialized field for natural language input, In the step of providing the above target conversation as input data, the agent, When the first target conversation is selected, the first target conversation is provided to the first model as the input data, and when the second target conversation is selected, the second target conversation is provided to the second model as the input data. Computer implementation method.

22. In paragraph 1, The above language model Stored in the above memory or stored in an external device connected to the conference service device via a network, and executed by the agent, Computer implementation method.

23. In paragraph 1, The above language model is, Generating the above output data in the form of at least one response selected from the group consisting of questions, answers, comments and summaries. Computer implementation method.

24. In paragraph 23, The above response format is, 1) The language model determines based on the target conversation, or 2) The above agent decides based on the above target conversation, Computer implementation method.

25. In paragraph 1, The step of providing the above conversation response is: Characterized in that the above conversation response is provided to the participant terminal in the form of a message by an AI (Artificial Intelligence) conference participant that is visually distinct from the conference participants. Computer implementation method.

26. In paragraph 1, The steps for providing the above conversation response are Deciding whether to provide the conversation response based on the access rights granted to each of the above meeting participants; Computer implementation method.

27. In paragraph 1, The steps for providing the above conversation response are Provides conversation responses obtained using local internal data with a security level corresponding to the access rights of each conference participant to each participant terminal of the above conference participants. Computer implementation method.

28. In paragraph 1, The steps for providing the above conversation response are Characterized in that the basis data used to obtain the output data among the local internal data and / or an access link to the basis data are provided together with the conversation response. Computer implementation method.

29. In paragraph 1, The step of providing the above conversation response is: If the target conversation is a conversation for which a response is individually requested by the participant terminal, the conversation response is provided only to the participant terminal, Computer implementation method.

30. In paragraph 1, A step in which the agent obtains a meeting result report including summary results for each of a plurality of preset criteria items from the conversations; the plurality of criteria items include at least one selected from the group consisting of a meeting overview item, a meeting content item, a main conversation collection item, a task item, and a next meeting item; The agent further comprises a step of providing the meeting result report to the participant terminal. Computer implementation method.

31. In paragraph 30, The above agent, 1) a step of scheduling the next meeting item based on the approval response of the meeting participant for the next meeting item; and / or 2) Further comprising a step of performing tracking management for the work item based on information about the work progress and / or work result received from the participant terminal associated with the work item. Computer implementation method.

32. In paragraph 1, Before receiving the above conversations, a step of receiving a meeting opening request including the meeting period, meeting content, and meeting participants information from the host terminal; A step of creating an online conference room including a conversation thread based on conversation input of the conference participants based on the conference opening request; and Further comprising a step of controlling the progress of the conference by non-real-time participation of the conference participants by providing the online conference room through the network in response to requests from each participant terminal of the conference participants during the conference period. Computer implementation method.

33. In paragraph 32, Further comprising a step of acquiring video data related to the above meeting content, The above online conference room further includes the above video data, Computer implementation method.

34. A language model that generates a natural language response using local internal data for natural language input and a memory storing at least one command; and A processor configured to execute at least one instruction stored in the memory, wherein the at least one instruction, when executed by the processor, causes the processor to perform the following method, the method comprising: The agent acquires conversations received from meeting participants participating in an online conference room that includes an agent connected to the above language model, The agent selects a target conversation from among the conversations, and the target conversation includes a conversation that requires a response using local internal data. The above agent provides the target conversation as input data to the language model, The above agent obtains output data for the input data from the language model, The agent provides a conversation response to the target conversation to the participant terminal of the meeting participant based on the output data. Conference service device.

35. A computer program stored in a computer-readable non-transitory recording medium, wherein the computer program includes instructions that, when executed by one or more processors, cause the one or more processors to perform a computer-implemented method, and the computer-implemented method comprises: A step in which the agent acquires conversations received from meeting participants participating in an online conference room including an agent connected to a language model; A step in which the agent selects a target conversation from among the conversations; the target conversation includes a conversation that requires a response using local internal data; A step in which the agent provides the target conversation as input data to the language model; wherein the language model generates a natural language response using the local internal data for the natural language input; A step in which the agent obtains output data for the input data from the language model; and The step of the agent providing a conversation response for the target conversation to the participant terminal of the conference participant based on the output data, A computer program stored on a computer-readable, non-transitory storage medium.

36. In a computer-readable non-transitory recording medium storing a computer program, the computer program includes instructions that, when executed by one or more processors, cause the one or more processors to perform a computer-implemented method, the computer-implemented method comprising: A step in which the agent acquires conversations received from meeting participants participating in an online conference room including an agent connected to a language model; A step in which the agent selects a target conversation from among the conversations; the target conversation includes a conversation that requires a response using local internal data; A step in which the agent provides the target conversation as input data to the language model; wherein the language model generates a natural language response using the local internal data for the natural language input; A step in which the agent obtains output data for the input data from the language model; and The step of the agent providing a conversation response for the target conversation to the participant terminal of the conference participant based on the output data, A computer-readable, non-transitory recording medium that stores a computer program.

Images