Method, apparatus and program for providing creator support service based on multi-agent artificial intelligence collaboration system

Abstract

A method for providing a creator support service based on a multi-agent artificial intelligence collaboration system according to various embodiments of the present invention is disclosed. The method may include: a step of obtaining a user query from a user account; a step of determining at least one worker agent corresponding to the user query through a master agent; a step of generating an expanded query by performing query expansion on the user query through the at least one worker agent; a step of obtaining result data corresponding to the user query by inputting the expanded query into a pre-trained large-scale language model (LLM); and a step of providing the result data to the user account.

Description

Technology Field

[0001] The present invention relates to a method, apparatus, and program for providing a creator support service based on a multi-agent artificial intelligence collaboration system. Background Technology

[0003] The creator economy refers to a massive digital ecosystem where individuals generate revenue based on their unique content and personalities. Hundreds of millions of people worldwide are participating in this new economic system, creating various added values ​​such as education, personal branding, and e-commerce through the formation of relationships with fans. The global creator economy is estimated to be worth hundreds of billions of dollars and is projected to show a high annual growth rate of over 10%; accordingly, the advertising market utilizing social media is also continuously growing.

[0004] Along with the growth of this market, there is an increasing demand for MCNs (Multi-Channel Networks) and personal branding support services to reduce the creative burden on creators and help them grow their brands. In particular, the advancement of artificial intelligence (AI) technology, especially generative AI and intelligent agent technology, is emerging as an important tool for creators to consistently publish content.

[0005] However, there is a need for improvement in current creator support technologies and personalized recommendation systems. Existing generative AI services sometimes tend to generate standardized results, which limits their ability to reflect a creator's unique personality or style, and makes it difficult to discover niche information that meets specific criteria, beyond general information widely known to the public.

[0006] Furthermore, existing recommendation systems sometimes struggle to provide optimized results by comprehensively reflecting the user's current location, temporal context, and detailed personal preferences in real time. This is because there are technical limitations in the systems' ability to systematically model and analyze complex relational data, such as location information, user behavior patterns, and social network interactions. Consequently, creators still face the inconvenience of having to spend a significant amount of time planning content themselves based on static recommendations provided by AI.

[0007] Therefore, there is a demand in the industry for advanced content production support methods. In this regard, Korean Published Patent No. 10-2023-0034758 discloses a creator support system, a creator support method, and a program stored on a recording medium. The problem to be solved

[0009] The present invention, devised in response to the aforementioned background technology, aims to provide a method, apparatus, and program for providing a creator support service based on a multi-agent artificial intelligence collaboration system.

[0010] The technical problems of the present invention are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0012] According to an embodiment of the present invention for solving the problem described above, a method for providing a creator support service based on a multi-agent artificial intelligence collaboration system is disclosed. The method may include: a step of obtaining a user query from a user account; a step of determining at least one worker agent corresponding to the user query through a master agent; a step of generating an expanded query by performing query expansion on the user query through the at least one worker agent; a step of obtaining result data corresponding to the user query by inputting the expanded query into a pre-trained large-scale language model (LLM); and a step of providing the result data to the user account.

[0013] In an alternative embodiment, the step of determining at least one worker agent corresponding to the user query through the master agent may include: extracting at least one entity name from the user query; extracting a user intent from the user query; extracting the conversation context of the current conversation from a conversation history containing the user query; and determining the at least one worker agent based on the at least one entity name, the user intent, and the conversation context.

[0014] In an alternative embodiment, the step of generating an expanded query by performing a query expansion for the user query through at least one worker agent may include: obtaining context data from pre-specified bookmark information in the user account or from the file system of the user terminal to which the user account is connected, based on a specific tag command included in the user query; obtaining real-time user data including the current location and current time of the user account, and obtaining real-time situation data corresponding to the real-time user data from an external API; and generating the expanded query by combining the user query, the context data, the real-time user data, and the real-time situation data.

[0015] In an alternative embodiment, the step of generating an expanded query by performing query expansion on the user query through the at least one worker agent may further include: extracting at least one keyword from the user query; searching for entities or relationships of past conversations associated with the at least one keyword from a knowledge base containing the past conversation history of the user account to obtain additional context data; and, if the additional context data is obtained, combining the user query, the context data, and the additional context data to generate the expanded query.

[0016] In an alternative embodiment, the step of generating an expanded query by performing query expansion on the user query through at least one worker agent may include: a step of extracting key keywords and relationships from the expanded query when the length of the expanded query exceeds a threshold; and a step of generating an optimal expanded query by summarizing and compressing the expanded query based on the key keywords and relationships.

[0017] In an alternative embodiment, the step of inputting the extended query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to the user query may include: obtaining persona information and output format information set in the user account; generating a prompt including the extended query, the persona information, and the output format information; and inputting the prompt into the large-scale language model to obtain the result data reflecting the persona information and the output format information.

[0018] In an alternative embodiment, the method may further include: a step of crawling a plurality of source data from the web to extract knowledge data including entities and relationships between said entities; a step of calculating weights for relationships between said entities based on a PageRank algorithm; a step of constructing a knowledge base used to obtain said extension query based on said knowledge data and said weights; a step of periodically searching the web for new entities associated with entities included in said knowledge base; and a step of, when said new entities are found, adding new knowledge data to said knowledge base including said new entities and relationships between said new entities and other entities previously stored therein.

[0019] In an alternative embodiment, the method may further include the steps of: generating a post based on the result data, scheduling the post to be published on a platform set in the social media user account, and providing an automatic response to comments on the post; providing performance indicators for posts published to date through the user account; searching a plurality of similar accounts that publish posts similar to the post from a knowledge base; and generating guidance information based on the content publishing pattern and engagement characteristics of a specific similar account among the plurality of similar accounts that has obtained higher performance indicators than the user account, and providing this information to the user account.

[0020] According to one embodiment of the present invention for solving the above-described problem, an apparatus is disclosed. The apparatus comprises: a memory for storing one or more instructions; and a processor for executing the one or more instructions stored in the memory, and the processor can perform the above-described methods by executing the one or more instructions.

[0021] According to one embodiment of the present invention for solving the above-described problem, a computer program stored on a recording medium readable by a computer is disclosed, which is combined with a computer, which is hardware, to perform the above-described methods.

[0022] Other specific details of the present invention are included in the detailed description and drawings. Effects of the invention

[0024] The present invention can provide an intelligent creator support service that supports the creation of customized content and provides data-based growth guidance by comprehensively analyzing a creator's personalized data, real-time situational information, and performance data of similar accounts.

[0025] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below. Brief explanation of the drawing

[0027] FIG. 1 is a drawing illustrating a system according to one embodiment of the present invention. FIG. 2 is a hardware configuration diagram of a computing device according to one embodiment of the present invention. FIGS. 3 to 8 are drawings illustrating a method for providing a creator support service based on a multi-agent artificial intelligence collaboration system according to an embodiment of the present invention. Specific details for implementing the invention

[0028] Various embodiments are now described with reference to the drawings. In this specification, various descriptions are provided to facilitate an understanding of the invention. However, it is evident that these embodiments can be practiced without such specific descriptions.

[0029] As used herein, terms such as “component,” “module,” “system,” etc. refer to computer-related entities, hardware, firmware, software, combinations of software and hardware, or executions of software. For example, a component may be, but is not limited to, a procedure executed on a processor, a processor, an object, an execution thread, a program, and / or a computer. For example, both an application executed on a computing device and the computing device itself may be a component. One or more components may reside within a processor and / or an execution thread. A component may be localized within a single computer. A component may be distributed among two or more computers. Additionally, these components may be executed from various computer-readable media having various data structures stored therein. Components may communicate through local and / or remote processes, for example, according to signals having one or more data packets (e.g., data from a component interacting with another component in a local system or distributed system, and / or data transmitted through signals to other systems and networks such as the Internet).

[0030] Furthermore, the term "or" is intended to mean an implicit "or" rather than an exclusive "or." That is, unless otherwise specified or evident from the context, "X uses A or B" is intended to mean one of the natural implicit substitutions. In other words, if X uses A; if X uses B; or if X uses both A and B, "X uses A or B" may apply to any of these cases. Additionally, the term "and / or" as used herein should be understood to refer to and include all possible combinations of one or more of the enumerated related items.

[0031] Additionally, the terms “comprising” and / or “comprising” should be understood to mean that such features and / or components are present. However, the terms “comprising” and / or “comprising” should be understood not to exclude the presence or addition of one or more other features, components and / or groups thereof. Furthermore, unless otherwise specified or clearly evident from the context to indicate a singular form, the singular in this specification and claims should generally be interpreted to mean “one or more.”

[0032] Those skilled in the art should recognize that the various exemplary logical blocks, configurations, modules, circuits, means, logics, and algorithmic steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or a combination of both. To clearly exemplify the interchangeability of hardware and software, various exemplary components, blocks, configurations, means, logics, modules, circuits, and steps have been generally described above in terms of their functionality. Whether such functionality is implemented in hardware or software depends on the specific application and design constraints imposed on the overall system. Skilled technicians may implement the described functionality in various ways for each specific application. However, such decisions regarding implementation should not be interpreted as moving out of the scope of the invention.

[0033] The description of the presented embodiments is provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art. The general principles defined herein may be applied to other embodiments without departing from the scope of the present invention. Thus, the present invention is not limited to the embodiments presented herein. The present invention should be interpreted in the broadest possible scope consistent with the principles and novel features presented herein.

[0034] In this specification, the term "computer" refers to any type of hardware device comprising at least one processor, and may be understood to include software configurations operating on said hardware device according to the embodiments. For example, the term "computer" may be understood to include smartphones, tablet PCs, desktops, laptops, and user clients and applications running on each of these devices, but is not limited thereto.

[0035] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0036] Each step described in this specification is described as being performed by a computer, but the subject of each step is not limited thereto, and depending on the embodiment, at least some of each step may be performed on different devices.

[0038] FIG. 1 is a drawing illustrating a system according to one embodiment of the present invention.

[0039] Referring to FIG. 1, a system according to one embodiment of the present invention may include a computing device (100), a user terminal (200), and an external server (300). The system illustrated in FIG. 1 is according to one embodiment, and its components are not limited to the embodiment illustrated in FIG. 1 and may be added, changed, or deleted as needed.

[0040] In one embodiment, the computing device (100) can provide a creator support service based on a multi-agent artificial intelligence collaboration system. For example, the computing device (100) can provide a comprehensive management service to a creator, including the creation of personalized content drafts, customized location recommendations, and data-based growth guidance, through the collaboration of a plurality of specialized artificial intelligence agents.

[0041] Specifically, the computing device (100) can obtain user queries from a user account. Here, the user account may include the account of a creator who generates revenue based on their own content and personality on social media, etc. And, the user query may refer to a request that the creator inputs in natural language to an AI assistant to perform specific tasks, such as content creation, information search, or location recommendation.

[0042] Additionally, the computing device (100) can determine at least one worker agent that responds to a user query through a master agent. Here, the master agent may refer to a central control agent that analyzes the intent of the user query, coordinates the entire workflow, and delegates tasks to the worker agent best suited for each task. And, at least one worker agent may include a worker agent (sub-agent) designed to perform tasks in a specific area of ​​expertise, such as content drafting, place recommendations, review course design, and ad campaign matching.

[0043] Additionally, the computing device (100) can generate an expanded query by performing query expansion on a user query through at least one worker agent. Here, query expansion may refer to a process of reconstructing a query by adding the user's personal data (e.g., bookmarks, past conversation history, etc.), real-time contextual information, related keywords, etc., to the user's original query so that a large-scale language model can generate more accurate and richer results.

[0044] Additionally, the computing device (100) can input an extended query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to a user query. And, the computing device (100) can provide the result data to a user account.

[0045] For example, when a user asks a computing device (100) to "write a review post of a nearby restaurant," the device can generate and provide a draft of an Instagram post that comprehensively reflects the user's current location information, past restaurant preferences (long-term memory), and a designated persona style.

[0046] Accordingly, the computing device (100) of the present invention can provide a personalized intelligent support service that goes beyond simple content creation and understands and reflects the creator's unique identity and current situation. Accordingly, through the intelligent support service according to the present invention, the user can reduce the burden of continuously publishing content, systematically develop their SNS brand through interaction with an AI assistant, and expand the influence of their account by receiving a data-based growth guide.

[0047] Hereinafter, an example of a method in which a computing device (100) provides a creator support service based on a multi-agent artificial intelligence collaboration system is described with reference to FIGS. 3 to 8.

[0048] In various embodiments, the computing device (100) may provide Web or Application-based services. However, it is not limited thereto.

[0049] The computing device (100) may include any type of computer system or computer device, such as, for example, a microprocessor, a mainframe computer, a digital processor, a portable device, and a device controller. However, it is not limited thereto.

[0050] Hereinafter, the hardware configuration of the computing device (100) will be described with reference to FIG. 2.

[0051] Meanwhile, the user terminal (200) may be connected to the computing device (100) via the network (400) and may be a terminal of a user who uses the multi-agent artificial intelligence collaboration system-based creator support service provided by the computing device (100). For example, the user terminal (200) may include a user who wishes to receive support for content creation through an AI assistant, a user who wishes to analyze the performance of their social media account and receive guidance on growth direction, or a user who wishes to receive recommendations for or execute advertising campaigns suitable for their brand.

[0052] That is, the user terminal (200) may include a user terminal using a user account that provides a multi-agent artificial intelligence collaboration system-based creator support service to the computing device (100) of the present invention.

[0053] Here, the user terminal (200) may include, for example, various types of computer devices. Specifically, for example, the user terminal (200) may refer to various terminal devices such as smartphones, tablet PCs, desktops, and laptops.

[0054] The user terminal (200) includes a display on at least a part of the terminal and may include an operating system for running applications or extension-based services provided by the computing device (100). For example, the user terminal (200) may be a smartphone, but is not limited thereto, and the user terminal (200) may include all kinds of handheld-based wireless communication devices such as navigation, PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartpad, tablet PC, etc., as wireless communication devices that ensure portability and mobility.

[0055] An external server (300) can be connected to a computing device (100) through a network (400), and can transmit and receive various information / data necessary for the computing device (100) to provide various functions, technologies, or services related to the method of the present invention, and can store and manage various information / data generated as the computing device (100) provides various functions, technologies, or services related to the method of the present invention.

[0056] For example, the external server (300) may be a database server that stores information used in various functions, technologies, or services related to the method of the present invention. As another example, the external server (300) may be a server that provides information used in various functions, technologies, or services related to the method of the present invention.

[0057] The network (400) may refer to a connection structure capable of exchanging information between each node, such as computing devices, multiple terminals, and servers. For example, the network (400) includes a Local Area Network (LAN), a Wide Area Network (WAN), the World Wide Web (WWW), a wired and wireless data network, a telephone network, a wired and wireless television network, etc.

[0058] Wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), 5GPP (5th Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), RF (Radio Frequency), Bluetooth network, NFC (Near-Field Communication) network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, etc.

[0060] FIG. 2 is a hardware configuration diagram of a computing device according to one embodiment of the present invention.

[0061] Referring to FIG. 2, a computing device (100) according to one embodiment of the present invention may include one or more processors (110), a memory (120) for loading a computer program (151) executed by the processor (110), a bus (130), a communication interface (140), and a storage (150) for storing the computer program (151). Here, FIG. 2 illustrates only the components related to the embodiment of the present invention. Therefore, a person skilled in the art to which the present invention pertains will understand that other general-purpose components may be included in addition to the components illustrated in FIG. 2.

[0062] The processor (110) controls the overall operation of each component of the computing device (100). The processor (110) may be composed of one or more cores and may include processors for data analysis and deep learning, such as a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), a neural processing unit (Neural Processing Unit), or a tensor processing unit (TPU) of the computing device. Alternatively, it may be configured to include any type of processor well known in the art of the present invention.

[0063] Additionally, the processor (110) can perform operations for at least one application or program for executing the method according to embodiments of the present invention, and the computing device (100) may have one or more processors.

[0064] In various embodiments, the processor (110) may further include RAM (Random Access Memory, not shown) and ROM (Read-Only Memory, not shown) for temporarily and / or permanently storing signals (or data) processed within the processor (110). Additionally, the processor (110) may be implemented in the form of a system-on-chip (SoC) comprising at least one of a graphics processing unit, RAM, and ROM.

[0065] Memory (120) stores various data, instructions and / or information. Memory (120) may load a computer program (151) from storage (150) to execute a method / operation according to various embodiments of the present invention. When the computer program (151) is loaded into memory (120), the processor (110) may perform the method / operation by executing one or more instructions constituting the computer program (151). Memory (120) may be implemented as volatile memory such as RAM, but the technical scope of the present invention is not limited thereto.

[0066] The bus (130) provides communication functions between components of the computing device (100). The bus (130) can be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

[0067] The communication interface (140) supports wired and wireless internet communication of the computing device (100). Additionally, the communication interface (140) may support various communication methods other than internet communication. To this end, the communication interface (140) may be configured to include a communication module well known in the art of the present invention. In some embodiments, the communication interface (140) may be omitted.

[0068] Storage (150) can store a computer program (151) non-temporarily. When performing a process according to an embodiment of the present invention through a computing device (100), storage (150) can store various information necessary to perform a method according to the disclosed embodiment or to provide a service.

[0069] The storage (150) may be configured to include non-volatile memory such as ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), flash memory, a hard disk, a removable disk, or any form of computer-readable recording medium well known in the art to which the present invention belongs.

[0070] A computer program (151) may include one or more instructions that cause a processor (110) to perform a method / operation according to various embodiments of the present invention when loaded into memory (120). That is, the processor (110) may perform the method / operation according to various embodiments of the present invention by executing the one or more instructions.

[0071] In one embodiment, the computer program (151) may include one or more instructions to perform various methods related to various tasks related to learning a neural network model.

[0072] The steps of the method or algorithm described in connection with embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or implemented by a combination thereof. The software module may reside in RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, a hard disk, a removable disk, a CD-ROM, or any form of computer-readable recording medium well known in the art to which the present invention belongs.

[0073] The components of the present invention may be implemented as a program (or application) and stored on a medium to be executed in combination with a computer, which is hardware. The components of the present invention may be implemented as software programming or software elements, and similarly, embodiments may be implemented in programming or scripting languages ​​such as C, C++, Java, assembler, etc., including various algorithms implemented as combinations of data structures, processes, routines, or other programming configurations. Functional aspects may be implemented as algorithms executed on one or more processors.

[0075] FIGS. 3 to 8 are drawings illustrating a method for providing a creator support service based on a multi-agent artificial intelligence collaboration system according to an embodiment of the present invention.

[0076] Referring to FIG. 3, the computing device (100) can obtain a user query from a user account (S110).

[0077] Specifically, the computing device (100) can receive a request for a creator support service, i.e., a user query, from a user terminal (200) via a network (400).

[0078] Here, the user terminal (200) may be a smartphone, tablet, or personal computer running a mobile application in which the service according to the present invention is installed, and the user account may include an account connected to the user terminal (200) linked to the creator's social media account.

[0079] User queries in the present invention are not limited to simple text input and may include commands or questions in the form of natural language that a creator inputs into a conversation window with an AI assistant. Additionally, user queries may include multimodal input combining images and natural language, or input of a URL (Uniform Resource Locator) for a specific location, thereby enabling richer and more intuitive service requests.

[0080] For example, the computing device (100) can obtain the photo data and text data as a single multimodal user query when a creator uploads a photo taken at a specific location through a mobile application installed on a user terminal (200) and simultaneously enters text such as 'write a draft of an Instagram post that matches the mood of this photo' into an AI assistant conversation window.

[0082] In one embodiment, the computing device (100) can determine at least one worker agent that responds to a user query through a master agent (S120).

[0083] Referring to FIG. 4, the computing device (100) can extract at least one entity name from a user query (S121).

[0084] Specifically, the computing device (100) can analyze a user query using a Named Entity Recognition (NER) module or a NER tool. Here, the NER module can identify the sentence structure and vocabulary of the user query based on a pre-trained natural language processing model, and identify and classify key named entities of predefined types, such as places, people, objects, times, and purposes, within the user query. This can be applied not only to text but also to multimodal queries entered by the user along with images.

[0085] For example, when a user inputs a query such as “Recommend a good pasta restaurant in Seongsu-dong to eat dinner with a friend,” the computing device (100) can extract and tag ‘Seongsu-dong’ as a ‘Location’, ‘friend’ as a ‘Companion’, ‘dinner’ as a ‘Time’, ‘pasta’ as a ‘Food’, and ‘restaurant’ as a ‘Purpose’.

[0086] Additionally, the computing device (100) can extract user intent from user query (S122).

[0087] Specifically, the computing device (100) can perform intent classification to determine the overall meaning and goal of a user query in parallel or sequentially with named entity recognition. More specifically, the computing device (100) can classify the user query into one of predefined intent classes, such as 'content creation', 'place recommendation', 'schedule design', etc., by analyzing the verb phrases, key vocabulary, context, etc. of the user query based on a Natural Language Understanding (NLU) algorithm. Here, each intent class can be directly mapped to the function of a specific worker agent.

[0088] For example, the computing device (100) can classify a query such as “recommend a restaurant in Seongsu-dong” as having a “place recommendation” intention based on the predicate “recommend”. As another example, the computing device (100) can clearly distinguish a query such as “write a draft of a post reviewing a restaurant in Seongsu-dong” as having a “content creation” intention based on the expression “write a draft of a post”.

[0089] Additionally, the computing device (100) can extract the conversation context of the current conversation from a conversation history containing user queries (S123).

[0090] Specifically, the computing device (100) may manage the context through a conversation context management module or a memory agent to ensure the continuity of the conversation and generate a personalized response. Here, the conversation context may consist of an in-memory based short-term memory that remembers the flow of the current conversation session and a long-term memory that stores user preferences or key information accumulated from multiple past conversation sessions in a structured format, such as in a graph database.

[0091] For example, the computing device (100) can extract from 'short-term memory' the content that the user said a few minutes ago, “I plan to go to Seongsu-dong this weekend,” and retrieve from 'long-term memory' the content that the user said a few weeks ago, “I like Italian food.” Subsequently, when the user asks, “Recommend a place to eat there,” the computing device (100) can identify through short-term memory that “there” is “Seongsu-dong” and identify through long-term memory that the user’s food preference is “Italian,” and synthesize this into the context of the current conversation.

[0092] And, the computing device (100) can determine at least one worker agent based on at least one entity name, user intent and conversation context (S124).

[0093] Specifically, the master agent of the computing device (100) can comprehensively analyze all information extracted in the preceding steps (i.e., entities, intents, and context) and route the task to the worker agent that can best perform the task. Here, routing can be performed by a predefined rule-based system or by a policy model trained to select the optimal agent through reinforcement learning. In this process, the master agent may establish an execution plan to sequentially call multiple worker agents.

[0094] For example, the computing device (100) can recognize that the task is to ‘recommend a personalized place’ by combining information such as the entity name {place: ‘Seongsu-dong’, food: ‘pasta’}, the intent ‘place recommendation’, and the conversation context {preference: ‘Italian’}. Accordingly, the master agent can determine a ‘PlaceRecommendationAgent’ that specializes in performing the place recommendation function among various worker agents, and instruct the agent to perform the next step by transmitting all extracted information to the agent.

[0095] For example, referring to FIG. 5, at least one worker agent of the present invention may include a plurality of worker agents that professionally perform specific tasks delegated by a master agent, and tools used by them.

[0096] Specifically, the worker agent may include a Place Recommendation Agent that recommends personalized places by comprehensively analyzing the user's past visit history, preferences, and current location; a Place Mining Agent that extracts place names from user queries and collects normalized place information by linking with external map APIs; a Persona Writer Agent that generates content drafts tailored to the persona style set by the user; a Scheduling Agent that designs a review course including the optimal visit route when multiple points of interest are given; a Campaign Matching Agent that recommends the optimal advertising campaign by calculating the similarity between the creator and the advertiser; and a Memory Agent that manages the short-term and long-term memory of conversations to provide necessary context to other agents.

[0097] In addition, the above worker agents may utilize tools such as NER / GraphSearch Tool, which recognizes key entities (NER) in user queries or converts natural language into graph database queries, and HybridRetrieval Tool, which combines keyword search and vector similarity search to improve search accuracy, in order to perform their tasks.

[0098] Through these tools, the present invention can efficiently perform complex multidimensional relationship analysis that is difficult to process in existing relational databases by utilizing a natural language-to-graph query conversion technology based on a fine-tuned NL2Cypher LLM model and a hybrid search system that considers both keywords and similarity. Accordingly, the present invention can realize multidimensional personalized recommendations that comprehensively consider user preferences, geographical constraints, and temporal context by analyzing multilayered correlations between user, content, and location in real time.

[0099] In various embodiments, the worker agents can perform complex and dynamic tasks more efficiently by utilizing additional tools such as a planning tool that establishes an execution plan for agents and tools to be used based on analyzed user intent and receives user feedback, an instruction retrieval tool that searches for past successful execution history to skip the planning step and quickly derive results, an execution tool that actually runs agents and tools according to the established plan and manages the session state, and an instruction memory tool that stores successful execution history in a database so that it can be reused.

[0100] Meanwhile, the expanded query generated by the aforementioned worker agents can be input into the LLM to generate result data.

[0102] In one embodiment, the computing device (100) can generate an expanded query by performing query expansion on a user query through at least one worker agent (S130).

[0103] Referring to FIG. 6, the computing device (100) can obtain context data from a pre-specified bookmark information in a user account or from the file system of a user terminal connected to the user account based on a specific tag command included in a user query (S131).

[0104] Specifically, the computing device (100) can perform the function of recognizing a specific predefined tag command (e.g., '@') so that the user can directly utilize their knowledge or data in the AI's work.

[0105] For example, when a user inputs a query including a specific tag command into a conversation window with an AI assistant, the computing device (100) can search for the document indicated by the command from the user account's personal storage space, 'My Library', and load the relevant content as context data.

[0106] For example, when a user queries, "Compare and analyze '@CafeAReview' and '@CafeBReview' and write a post," the computing device (100) recognizes the '@' tag, searches for the full text of the 'CafeAReview' document and the 'CafeBReview' document stored in 'My Archive' respectively, and obtains them as context data so that they can be referenced in subsequent operations.

[0107] As another example, when a user inputs a specific tag command indicating a local file system, such as “@localfile:plan.pdf’ summarize,” the computing device (100) may recognize this as a command to access a file stored in the local storage of the user terminal (200). In this case, an application running on the user terminal (200) may obtain file access permission from the user, read the entire text of the ‘plan.pdf’ file, and transmit it to the computing device (100) via the network (400), thereby allowing the computing device (100) to obtain context data.

[0108] Additionally, the computing device (100) can obtain real-time user data including the current location and current time of the user account, and obtain real-time situation data corresponding to the real-time user data from an external API (S132).

[0109] Specifically, the computing device (100) can obtain real-time user data including the user's current location coordinates and current time information from a sensor (e.g., GPS) or operating system of the user terminal (200) to generate results optimized for the time and place of the query. Subsequently, the computing device (100) can collect additional information related to the current situation as real-time situation data by calling various external APIs provided by an external server (300) using the obtained real-time user data as a key value. Here, the external APIs may include APIs that provide weather information, regional event information, location congestion information, traffic information, etc.

[0110] For example, the computing device (100) can obtain real-time user data that the user's current location is 'Seongsu-dong' and the current time is '2 PM'. Subsequently, the computing device (100) can obtain real-time situation data by transmitting the 'Seongsu-dong' location value to an external map API to obtain 'nearby pop-up store event information', and by transmitting the '2 PM' value to an external weather API to obtain weather information such as 'sunny, 25 degrees'.

[0111] And, the computing device (100) can generate an extended query by combining user query, context data, real-time user data and real-time situation data (S133).

[0112] Specifically, the computing device (100) can integrate all information obtained in the preceding steps into a structured form so that the large-scale language model (LLM) can generate richer and more accurate results in subsequent steps. More specifically, the computing device (100) can go beyond simple text combination and clearly distinguish the source and type of each data (original query, custom document, real-time location, external situation, etc.) to reconstruct it into a form (e.g., a prompt object in JSON format) that allows the LLM to better understand the context.

[0113] For example, the computing device (100) can generate an extended query by structuring each piece of information into clear key and value, such as 'Original query: Cafe comparison post', 'Reference material 1: Body of Cafe A review...', 'Current location: Seongsu-dong', 'Current weather: Sunny, 25 degrees', 'Nearby information: Pop-up store event in progress'. The extended query generated in this way is delivered to the LLM and can serve as a basis for generating results that go beyond simply comparing two cafes and add creative context, such as 'enjoying a pop-up store together in Seongsu-dong on a sunny day'.

[0115] In various embodiments, the computing device (100) may obtain additional context based on a knowledge base containing the past conversation history of a user account and reflect this in an extended query.

[0116] For example, when a user’s query is abstract or lacks specific conditions, such as “recommend what to do today,” the computing device (100) can identify the user’s hidden intentions or preferences by referring to past conversation history and provide personalized results based on this.

[0117] Specifically, the computing device (100) can extract at least one keyword from a user query.

[0118] More specifically, the computing device (100) can analyze the currently input user query based on natural language processing and identify and extract keywords such as nouns and verbs that play a key role in constructing the meaning of the sentence. To this end, the computing device (100) can evaluate the importance of words within the query through statistical techniques or a pre-trained language model.

[0119] For example, when a user asks, "Recommend a quiet cafe to go to this weekend," the computing device (100) can extract keywords such as 'weekend', 'quiet cafe', 'recommendation', etc.

[0120] And, the computing device (100) can obtain additional context data by searching for entities or relationships of past conversations associated with at least one keyword from a knowledge base containing the past conversation history of a user account.

[0121] Specifically, the computing device (100) can search a knowledge base in which long-term memory is stored by using the previously extracted keyword as an entry point through a MemoryAgent.

[0122] The knowledge base of the present invention may include a graph database (Graph DB) in which entities and relationships extracted from past conversations are structured and stored in a graph form. That is, the computing device (100) can go beyond simply finding past conversations containing keywords and can search for structured information such as user preferences and past behavior patterns that are directly or indirectly connected to the keywords.

[0123] For example, the computing device (100) can search a knowledge base in which long-term memory is stored using the extracted keyword 'cafe'. Additionally, the computing device (100) can search for relationship data in which the user showed a positive response to entities such as 'acidic coffee' and 'bakery' in past conversations. Accordingly, the computing device (100) can obtain information on the user's potential preferences, such as 'acidic coffee' and 'bakery', as additional context data.

[0124] Meanwhile, when the computing device (100) obtains additional context data, it can generate an extended query by combining the user query, the context data, and the additional context data.

[0125] Specifically, the computing device (100) can integrate all pieces of information collected so far into a single structured data so that in a subsequent step, a large-scale language model (LLM) can understand both the explicit and implicit requirements of the user and generate an optimal result. In this process, different weights may be assigned according to the source of each data (original query, user-defined bookmark, past conversation history, etc.), or specific roles may be assigned to create prompt objects.

[0126] For example, the computing device (100) can generate a final extended query by structuring each piece of information in JSON format with a clear role, such as 'Original query: "Quiet cafe recommendation", User provided context: "Body of a post related to minimalist interior...", User preference (long-term memory): "Acidic coffee"'. The extended query generated in this way can serve as a basis for recommending places that are more satisfying to the user by conveying to the LLM not only the user's explicit requirements but also the reference materials provided directly and potential preferences accumulated in the past.

[0128] In various embodiments, the computing device (100) can perform a process of summarizing and compressing the extended query and accumulated conversation context when they exceed the context window limit of the large language model (LLM) or become long enough to impair response quality.

[0129] Specifically, the computing device (100) can extract key keywords and relationships from the extended query when the length of the extended query exceeds a threshold.

[0130] More specifically, the computing device (100) may call a memory agent based on the judgment of a worker agent (e.g., PersonaWriterAgent). The called memory agent may use graph-based text summarization algorithms and / or topic modeling techniques, such as LLM-based summarization and TextRank, to identify and extract the most important keywords for constructing the meaning of sentences throughout a long extended query and the semantic relationships between them. Through this process, the computing device (100) may remove unnecessary or redundant information from the long extended query.

[0131] For example, when a computing device (100) generates an extended query of thousands of tokens by combining a user's bookmarks, conversation history, etc., it can extract key keywords such as 'Seongsu-dong', 'cafe', 'minimalism', 'acidic coffee' and key relationships such as '(Seongsu-dong)-[location]-(cafe)', '(cafe)-[feature]-(minimalism)' from the entire text through a memory agent.

[0132] And, the computing device (100) can generate an optimal extended query that summarizes and compresses the extended query based on key keywords and relationships.

[0133] Specifically, the computing device (100) can reconstruct the previously extracted core keywords and relationships so that their semantic structure is maintained, thereby generating a new query that is semantically identical to the original extended query but shorter in length. This process can be performed using a conditional generation model that must include the extracted keywords and relationships, or through a language model specialized for summarizing tasks.

[0134] For example, the computing device (100) can combine extracted keywords and relationships such as ‘Seongsu-dong’, ‘cafe’, ‘minimalism’, and ‘acidic coffee’ to generate an optimal expanded query that preserves the original intent while reducing the length, such as “Describe a cafe with minimalist interior located in Seongsu-dong. The cafe is characterized by acidic coffee.” This optimal expanded query can be efficiently processed within the context window of the LLM and prevents a decrease in response quality due to unnecessary information.

[0136] In one embodiment, the computing device (100) can input an extended query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to a user query (S140).

[0137] Referring to FIG. 7, the computing device (100) can obtain persona information and output format information set in a user account (S141).

[0138] Specifically, the computing device (100) may refer to a style guide pre-set by the user so that the content to be generated consistently maintains the creator's unique brand identity. Here, persona information may be information that defines the overall tone and manner, style, tone, etc. of the content. And, output format information may be information that defines the structure (e.g., paragraph composition, character limit, whether to include hashtags, etc.) suitable for the platform or purpose on which the content will be published.

[0139] For example, the computing device (100) can obtain, from the account settings of a specific travel creator, ‘a friendly and detailed guide style of a travel expert’ as persona information and ‘an Instagram post containing three paragraphs and five recommended hashtags’ as output format information.

[0140] Additionally, the computing device (100) can generate a prompt including an extension query, persona information, and output format information (S142).

[0141] Specifically, the computing device (100) can configure a prompt, which is a final work instruction to be delivered to a large-scale language model (LLM). Here, the prompt can serve to clearly indicate to the LLM what (extension query), in what style (persona information), and in what format (output format information) the result should be generated. Such a prompt can be generated by inserting each piece of information into a structured template.

[0142] For example, the computing device (100) can generate a final prompt by inserting the persona information of ‘travel expert style’ obtained in previous steps, the extension query of ‘information related to Seongsu-dong pop-up store’, and the output format information of ‘Instagram posting format’ into a prompt template such as “‘You are a {persona}. Refer to the following {extension query} and write a post in accordance with the {output format}.’”

[0143] And, the computing device (100) can input a prompt into a large-scale language model to obtain result data that reflects persona information and output format information (S143).

[0144] Specifically, the computing device (100) can transmit the completed prompt to an inference engine of a large-scale language model built externally or internally via an API and an agent-to-agent communication protocol such as MCP, A2A, AG-UI. Accordingly, the large-scale language model can interpret the complex instructions of the received prompt and create new text that fits the persona and output format based on a given extension query. The text thus generated may be the final result data.

[0145] For example, the computing device (100) can obtain as result data a text of a review of a visit to a Seongsu-dong pop-up store written in an Instagram posting format that includes three paragraphs and five recommended hashtags, with a persona of ‘a friendly and detailed guide style of a travel expert’ applied, as a result of inputting the previously generated prompt into the LLM.

[0147] In one embodiment, the computing device (100) can provide result data to a user account (S150).

[0148] Specifically, the computing device (100) can transmit the generated result data to a mobile application running on a user terminal (200) via a network (400). The result data can be displayed on an AI editor screen within the application so that the user can immediately view, modify, or publish it.

[0149] For example, the computing device (100) can transmit a generated post draft (result data) to a user terminal (200) to provide a function that allows the user to review the draft on an AI editor screen, modify the desired parts, and then press a 'publish' button to post it to their social media immediately or scheduled for a specific time.

[0150] Accordingly, the computing device (100) of the present invention can reduce the creator's creative burden by providing a customized content draft that is immediately usable and reflects the creator's unique identity (persona) and the specific purpose of use (output format) of the content.

[0152] In various embodiments, the computing device (100) can generate a post based on result data, schedule the post to be published to a platform set in a social media user account, and provide an automatic response to comments on the post.

[0153] Specifically, the computing device (100) can convert result data modified and finalized by the user within the AI ​​editor into a post format that conforms to the API specifications of the social media platform. The user can set up immediate publication or scheduled publication by specifying a specific time for the post.

[0154] Additionally, the computing device (100) can periodically monitor comments newly created on a post after the post is published, and automatically generate and post responses to specific types of comments (e.g., location inquiries, price inquiries, etc.) through pre-set rules or natural language analysis.

[0155] For example, if a user sets a cafe review post completed in an AI editor to 'publish tomorrow at 2:00 PM,' the computing device (100) can automatically post the post to a linked Instagram account at that time. After posting, if a comment asking 'Where is this?' is posted, the computing device (100) can automatically reply based on the location information of the post, such as 'It is OOO Cafe in Seongsu-dong!'

[0156] Accordingly, the computing device (100) of the present invention can reduce the operational burden on the creator by automating the process from the creation of content to publication and basic communication.

[0158] In various embodiments, the computing device (100) may provide performance indicators for posts published to date through a user account.

[0159] Specifically, the computing device (100) can periodically call the API of the linked social media platform to collect performance data for each post published by the user. The performance indicators include quantitative data such as the number of 'likes', comments, shares, reach, impressions, and engagement rate, and the collected data can be stored and managed by mapping it to each post through a 'My Archive' function, etc.

[0160] For example, the computing device (100) can collect performance indicators such as '500 likes, 30 comments, and 3,000 reach' by calling the Instagram API at 10:00 AM today for a post published yesterday. These performance indicators are provided in the form of visual graphs or charts when the user clicks on the post in 'My Library', which can help the user intuitively understand the performance of their content.

[0161] In other words, through the 'My Archive' feature, users can not only systematically track and manage performance indicators for each post, but also analyze performance data to utilize it as foundational material (context) for AI to generate better content in the future.

[0162] Additionally, the computing device (100) can search for multiple similar accounts that publish posts similar to the post from the knowledge base.

[0163] Specifically, the computing device (100) can convert user account profile information, the subject of content primarily published, text style, etc. into vectors and use this as a query to identify accounts with high similarity among numerous other creator account information stored in a knowledge base (e.g., Vector DB) as similar accounts.

[0164] For example, the computing device (100) can calculate the cosine similarity between a feature vector of a user account and another account vector in the knowledge base. The closer the similarity value is to 1 (i.e., the more similar the directions the two vectors point to), the more similar the content styles or topics of the two accounts are, and the computing device (100) can identify accounts with a similarity score above a preset threshold (e.g., 0.8) as similar accounts.

[0165] As another example, the computing device (100) can calculate the similarity between vectors using Euclidean distance. Here, Euclidean distance may be a value that measures how physically close the endpoints of two vectors are in a multidimensional space. That is, the closer the Euclidean distance is to 0 (i.e., the closer the two vectors are located in space), the more similar the characteristics of the two accounts are, and the computing device (100) can identify accounts where this distance is less than or equal to a preset threshold (e.g., 0.2) as similar accounts.

[0166] Additionally, the computing device (100) may search for other accounts clustered into the same category or domain as similar accounts by referencing an ontology built within a knowledge base (e.g., Graph DB).

[0167] For example, if a user account primarily publishes content related to 'home baking', the computing device (100) can search a knowledge base based on the characteristics of the account and identify similar 'home baking' professional creator accounts as multiple similar accounts.

[0168] And, the computing device (100) can generate guidance information based on the content publishing pattern and engagement characteristics of a specific similar account that has obtained a higher performance indicator than the user account among a plurality of similar accounts, and provide it to the user account.

[0169] Specifically, the computing device (100) can select benchmark accounts among identified similar accounts that have higher specific performance indicators than user accounts, such as the engagement-to-follower ratio. Subsequently, the computing device (100) can analyze 'content publishing patterns,' such as their content topics, formats (images, videos, etc.), publication cycles, and time zones, and 'engagement characteristics,' such as the characteristics of content that elicited a high response or the method of comment communication, in order to identify the factors that led to the success of the benchmark accounts. Then, the computing device (100) can compare the analysis results with the user's current operating method to derive areas for improvement and generate guidance information in the form of natural language.

[0170] For example, the computing device (100) can discover a pattern in which benchmark accounts commonly receive the highest response when publishing ‘recipes in the form of short videos (Reels)’ on ‘weekend mornings.’ Meanwhile, if a user account mainly publishes image content on weekday evenings, the computing device (100) can generate and provide specific guidance information to the user account, such as, ‘Try publishing recipe content in the form of Reels on the next weekend morning. Increased engagement is expected.’

[0171] Therefore, the computing device (100) of the present invention can perform the role of an AI consultant that goes beyond simply providing performance data and presents practical growth strategies to creators through data-based competitive analysis.

[0173] According to various embodiments of the present invention, a computing device (100) can pre-build a knowledge base that AI agents can utilize based on web crawling.

[0174] The knowledge base of the present invention may include a graph database and a vector database.

[0175] Here, a graph database can refer to a database specialized in intuitively storing and navigating complex connections between data by representing entities as nodes and relationships between entities as edges.

[0176] Furthermore, a vector database can refer to a database specialized in storing high-dimensional embedding vectors generated by extracting semantic features from data such as text or images, and in searching for semantically similar data at high speed by measuring the similarity between vectors.

[0177] Figure 8 is a diagram illustrating an example of the process of building and expanding such a knowledge base.

[0178] Referring to FIG. 8, a computing device (100) can crawl multiple source data from the web and extract knowledge data including entities and relationships between entities (S210).

[0179] Specifically, the computing device (100) can collect large amounts of text and metadata from public web sources, such as social media platforms, blogs, and news sites, using a web crawler. Subsequently, the computing device (100) can apply natural language processing techniques, such as Named Entity Recognition (NER) and Relation Extraction, to the collected unstructured data to extract knowledge data in a structured form, which includes entities that are units with contextual meaning and semantic relationships connecting these entities.

[0180] For example, a computing device (100) can crawl millions of social media posts related to a specific area (e.g., Seongsu-dong), and then analyze each post to extract knowledge data including an entity called 'Seongsu Cafe', an entity called 'Celebrity', and a relationship called 'visited' connecting the two entities.

[0181] Additionally, the computing device (100) can calculate weights for relationships between entities based on the PageRank algorithm (S220).

[0182] Specifically, the computing device (100) can model all extracted entities and relationships into one large network graph. Subsequently, the computing device (100) can iteratively calculate the importance of each entity within the network through the PageRank algorithm. Finally, the computing device (100) can calculate a weight representing the importance of a relationship based on the importance scores of the entities connected to a specific relationship. Here, the computing device (100) can calculate the weight of the relationship by assigning a higher weight to relationships connected to entities with higher importance.

[0183] For example, the computing device (100) can calculate and assign a higher weight (e.g., 0.9) to the relationship '(Celebrity)-[Visited]→(Seongsu Cafe)' than to other general relationships when the 'Celebrity' entity in the network receives links from numerous other entities and the importance score is calculated to be high.

[0184] And, the computing device (100) can build a knowledge base used to obtain an extended query based on knowledge data and weights (S230).

[0185] Specifically, the computing device (100) can store the previously extracted knowledge data and the calculated weights in a graph database. Here, entities are stored as nodes and relationships as edges, and the calculated weights can be stored together as property values ​​of the corresponding edges. Additionally, the computing device (100) can store embedding vectors representing the meaning of each entity together as properties of the nodes to improve semantic search efficiency in subsequent steps.

[0186] For example, the computing device (100) stores 'celebrity' and 'Seongsu Cafe' as nodes, respectively, and 'visited' as an edge connecting the two nodes in a graph database, and at this time, '0.9' can be recorded together as the weight attribute of the corresponding edge.

[0188] In various embodiments, the computing device (100) can continuously expand the knowledge base through an active crawler.

[0189] Specifically, the computing device (100) can periodically search for new entities associated with entities included in a knowledge base from the web.

[0190] More specifically, the computing device (100) can run an active crawler that uses entities already stored in a knowledge base (e.g., creators in a specific field, specific regions) as seeds to prioritize exploring new web pages or social media accounts that are highly relevant to said entities. That is, the computing device (100) can efficiently discover new data that is highly relevant compared to random web crawling through the active crawler.

[0191] For example, if the computing device (100) has a large number of 'home baking' related creators stored in the knowledge base, the active crawler can discover new 'home baking' creators that are not yet in the knowledge base by prioritizing the exploration of other accounts that the creators follow or mention.

[0192] And, when a new entity is discovered, the computing device (100) can add new knowledge data to the knowledge base, including the new entity and the relationship between the new entity and other entities that have been stored.

[0193] Specifically, the computing device (100) can perform entity and relationship extraction for newly discovered entities (e.g., new creator accounts) in a manner similar to the step (S210) described above. In this process, the computing device (100) can identify the relationships that the new entities have with other entities within the existing knowledge base, classify and cluster them to generate new knowledge data, and then expand the scope of the knowledge graph by integrating this into the existing knowledge base.

[0194] For example, when the computing device (100) discovers a new 'home baking' creator account, it can analyze the hashtags primarily used by the account, other accounts mentioned, etc., to establish relationships with other entities within the 'home baking' cluster of the existing knowledge base. Subsequently, the computing device (100) can expand the knowledge graph by adding new knowledge data containing the new account (new entity) and newly established relationships to the knowledge base.

[0195] Accordingly, the computing device (100) of the present invention can build and operate a knowledge base that continuously reflects the latest information on the web, going beyond simply using a static database.

[0197] According to an additional embodiment of the present invention, the computing device (100) can convert a natural language query into a graph database query to precisely search for places with complex conditions and design an optimal visiting course that reflects the user's current situation and preferences.

[0198] Specifically, the computing device (100) can utilize a finely tuned large language model specialized in natural language processing (NLP) to efficiently navigate a graph database containing complex relational data such as places, reviews, and related destinations.

[0199] More specifically, the computing device (100) can search for entities that accurately satisfy the user's complex search conditions through an NL2Cypher LLM model that converts natural language queries into Cypher, which is a standard query language of a graph database. In addition, the computing device (100) can perform a function of recommending a visiting course that includes an optimal route by combining real-time situational information such as the user's current location, time, and weather with user preference information stored in long-term memory.

[0200] For example, when a user inputs a natural language query containing complex conditions such as "recommend a quiet cafe in Seongsu-dong that has parking and allows pets," the computing device (100) can convert this into a Cypher query such as MATCH (p:Place)-[:IN_AREA]->(:Area {name: 'Seongsu-dong'}) WHERE p.category = 'Cafe' AND p.parking = true AND p.pets_allowed = true AND p.noise_level = 'quiet' RETURN p using NL2Cypher LLM. Afterwards, the computing device (100) can execute the converted query in a graph database to obtain a list of cafes that satisfy all of the conditions. Additionally, when the user makes an additional inquiry saying, “Plan a 3-hour date course that includes that cafe and nearby souvenir shops,” the computing device (100) can design and provide a visiting course that connects the recommended cafe and nearby souvenir shops in an optimal route by taking into account the current weather (e.g., clear) and the user’s preference (e.g., cute souvenirs).

[0201] Therefore, the computing device (100) can go beyond simple keyword search, understand the user's complex and multi-layered requirements in their natural language, and explore relationships between vast amounts of data to provide precise and personalized place recommendations and actionable activity plans.

[0203] According to an additional embodiment of the present invention, the computing device (100) can provide a digital marketing function that connects a creator and a brand.

[0204] Specifically, the computing device (100) can arrange an optimal match between a brand (advertiser) that wants to conduct an advertising campaign and a creator who wants to utilize their social media influence, and can predict and provide expected campaign performance.

[0205] More specifically, the computing device (100) can run a Campaign Matching Agent. The agent can embed meta-information such as the creator's content style, primary interests, follower demographics, and the advertising campaign's target audience, brand identity, and campaign goals into high-dimensional vectors. Subsequently, the computing device (100) can calculate the similarity between the creator vector and the campaign vector within a vector database to measure the suitability and predict the expected return on advertising spend (ROAS) based on past performance data of similar campaigns. Here, similarity can be determined by various similarity measurement algorithms, including cosine similarity and minimum edit distance.

[0206] For example, when a vegan cosmetics brand targeting women in their 20s registers a new product promotion campaign, the computing device (100) can convert the campaign information into a vector and compare its similarity with numerous beauty creator vectors within a knowledge base. Then, through the comparison results, the computing device (100) can identify a specific creator who frequently publishes vegan product review content and has a high proportion of female followers in their 20s as the optimal candidate. Additionally, the computing device (100) can analyze the average engagement rate and ROAS data of similar cosmetics campaigns conducted by the creator in the past, provide specific predictions such as 'Estimated ROAS: 350%' to the advertiser, and propose a match.

[0207] Accordingly, the computing device (100) of the present invention can build an efficient digital marketing ecosystem that increases the satisfaction of both parties by recommending to advertisers the optimal creators who can maximize the probability of campaign success through data-based objective indicators, and by providing creators with monetization opportunities that match their brand value.

[0208] Additionally, the computing device (100) may provide a function to recommend a collaboration partner that can maximize synergy between creators and to support the joint content creation process.

[0209] Specifically, the computing device (100) can search for and recommend other creators from a knowledge base that can create synergy with the characteristics of a specific creator, in accordance with the request of that creator. Additionally, the computing device (100) can increase the likelihood of success in collaboration by supporting joint content planning and drafting between creators who have successfully collaborated.

[0210] More specifically, the computing device (100) analyzes the account characteristics (e.g., content topics, audience analysis, persona, etc.) of a creator who wishes to collaborate, and can identify a mutually complementary creator as a collaboration partner candidate who has high topic relevance and low audience overlap within the knowledge base. Subsequently, the computing device (100) analyzes both the personas and content styles of the two creators selected as collaboration targets, and can generate a joint content topic and script draft through LLM that can harmoniously reveal the personalities of the two.

[0211] For example, when a creator who mainly deals with 'home cooking recipes' requests collaboration, the computing device (100) can recommend a creator specializing in 'pretty dish reviews' as the optimal collaboration partner based on a knowledge base, who shares a similar audience but whose content area does not overlap.

[0212] Additionally, the computing device (100) can suggest specific joint content ideas, such as "year-end home party food recipes using new dishes," by considering both creators' personas (e.g., "kind cooking teacher" and "stylish designer") along with recommendations from collaboration partners. Furthermore, the computing device (100) can support the two creators in collaborating smoothly by generating and providing a draft of a joint posting script in which the "home cooking" creator writes the food recipe section and the "dish review" creator writes the plating and design tips section.

[0213] Therefore, the computing device (100) can go beyond supporting the growth of individual creators and activate a 'creator network' that organically connects creators to create new value.

[0215] According to an additional embodiment of the present invention, the computing device (100) may provide an ad copy inspection function. Here, the ad copy inspection function may mean a function of reviewing and identifying in advance whether result data or post drafts generated by a creator may violate relevant laws, such as the Medical Advertising Act or the Fair Labeling and Advertising Act.

[0216] Specifically, the computing device (100) may receive text generated by a user (e.g., result data or post draft). Subsequently, the computing device (100) may use natural language processing (NLP) technology to analyze specific expressions, phrases, or claims included in the text and compare them with an external regulatory database (DB) in which guidelines from external regulatory agencies, such as the Ministry of Food and Drug Safety (MFDS), are stored.

[0217] Through this process, the computing device (100) can automatically screen whether the text contains false and exaggerated expressions prohibited by the Medical Advertising Act, the Fair Advertising Act (Exaggerated Advertising Act), etc., or claims of unapproved efficacy and effects.

[0218] In addition, the computing device (100) can provide the screening results to a user account to induce the creator to recognize in advance any parts that may violate regulations and to correct them.

[0219] For example, the computing device (100) can identify that if a creator writes the phrase “I ate this product and my insomnia was cured” in a draft review of a health functional food, the expression “cured” is interpreted as guaranteeing the treatment of a disease and may violate the Medical Advertising Act. Then, the computing device (100) can highlight the phrase on the AI ​​editor and provide a warning message such as “Potential violation of regulations: Guaranteed disease treatment effect” and a correction suggestion such as “Helpful for symptom relief” to the user account.

[0220] Therefore, the computing device (100) can protect both the creator and the advertiser brand by minimizing in advance the legal risks that may arise from the creator not being familiar with complex advertising-related laws and by increasing the credibility of the advertising content.

[0222] According to an additional embodiment of the present invention, the computing device (100) may provide a data analysis function that performs trend and commercial area analysis based on social data.

[0223] Specifically, the computing device (100) can analyze vast amounts of social media data related to a specific region or commercial district to identify the latest content trends of the region, predict the consumption patterns of visitors, and thereby provide information that predicts the commercial value or return on investment of a specific location.

[0224] More specifically, the computing device (100) can analyze numerous social media posts tagged with specific location information built into a knowledge base. In this process, the computing device (100) can identify entities such as business types, menus, and brands that are frequently mentioned in a specific commercial district, and analyze users' engagement (e.g., likes, comments, etc.) and sentiment (e.g., positive / negative) regarding said entities. Additionally, the computing device (100) can identify visitors' flow paths and associated consumption patterns by analyzing other places visited by users who have visited a specific place through the exploration of relationships in a graph database.

[0225] For example, when a user who wants to open a new themed cafe in a specific commercial area (e.g., Yeonnam-dong) queries for 'Yeonnam-dong cafe trend analysis,' the computing device (100) can analyze social media data related to Yeonnam-dong stored in a knowledge base. In this process, the computing device (100) can discover that engagement with recent dessert-related posts, such as 'yakgwa' and 'tanghulu,' has surged, and that visitors mainly visit 'prop shops' and cafes together. Based on these analysis results, the computing device (100) determines that a 'cafe with a prop shop concept selling desserts' has a high probability of success in that location and can generate and provide a commercial area analysis report that recommends the optimal location along with an expected return on investment based on social data.

[0226] Accordingly, the computing device (100) of the present invention can provide objective and quantitative analysis information based on vast social data, rather than intuition or limited information, to small business owners or prospective entrepreneurs.

[0228] According to an additional embodiment of the present invention, the computing device (100) can provide an intelligent content review and optimization function that predicts in advance the expected performance of a draft of content that a creator intends to publish and proposes an optimization method to maximize engagement.

[0229] Specifically, the computing device (100) receives a post draft (text, image, hashtag, etc.) and expected publication time information from a user account, and can predict the expected engagement (e.g., likes, comments, etc.) that can be obtained when the post is published as a quantitative figure. In addition, the computing device (100) can generate a content revision that can increase engagement based on the predicted performance and provide it to the user account.

[0230] More specifically, the computing device (100) can embed the text and images of the input post draft into vectors and search for semantically similar past posts within a knowledge base (e.g., Vector DB). Subsequently, the expected engagement score of the current draft can be calculated through a machine learning-based prediction model that analyzes the correlation between the actual performance indicators of the searched past posts and the publication time (day of the week, time of day), keywords used, and hashtags. Additionally, based on the analysis results of the prediction model and the latest trend information from the knowledge base (e.g., Graph DB), the computing device (100) can use LLM to generate specific modification suggestions (e.g., changing hashtags, modifying writing style, adding questions, etc.) regarding which elements of the current draft need to be modified to increase engagement.

[0231] For example, when a user submits a draft post about 'Jeju Island Cafe' to be published on Monday morning, the computing device (100) can analyze past data and provide a prediction such as "Expected engagement score: 65 / 100". Along with the prediction, the computing device (100) can make a time optimization suggestion such as "Based on performance analysis results, posts related to 'Jeju Island Cafe' generally have 30% higher engagement on Friday evenings. We recommend changing the publication time."

[0232] As another example, the computing device (100) can provide specific and actionable modification guides, such as “The current draft is focused on conveying information. Try adding a question like ‘Where is your favorite cafe in Jeju Island?’ to the last paragraph to encourage comment participation,” or “The exposure of the hashtag ‘#JejuLife’ has been high recently. Consider adding it to the existing hashtags.”

[0233] Therefore, the computing device (100) can help the creator publish content with a 'high probability of success' by predicting performance based on data and optimizing the content before publication, moving away from the passive method of publishing content and waiting for results based only on their experience or intuition.

[0235] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without altering its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive.

Claims

Claim 1 A method performed by a computing device comprising at least one processor, comprising: obtaining a user query from a user account; determining at least one worker agent corresponding to the user query through a master agent; generating an expanded query by performing query expansion on the user query through the at least one worker agent; inputting the expanded query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to the user query; and providing the result data to the user account; wherein the step of generating an expanded query by performing query expansion on the user query through the at least one worker agent comprises: obtaining context data from pre-specified bookmark information in the user account or from the file system of a user terminal connected to the user account based on a specific tag command included in the user query; obtaining real-time user data including the current location and current time of the user account, and obtaining real-time situation data corresponding to the real-time user data from an external API; and generating the expanded query by combining the user query, the context data, the real-time user data, and the real-time situation data. The step of generating an expanded query by performing query expansion on the user query through at least one worker agent, comprising: extracting at least one keyword from the user query; searching for entities or relationships of past conversations associated with the at least one keyword from a knowledge base containing the past conversation history of the user account to obtain additional context data; and, if the additional context data is obtained, generating the expanded query by combining the user query, the context data, and the additional context data.A method for providing a creator support service based on a multi-agent artificial intelligence collaboration system, further including Claim 2 A method performed by a computing device comprising at least one processor, comprising: obtaining a user query from a user account; determining at least one worker agent corresponding to the user query through a master agent; generating an expanded query by performing query expansion on the user query through the at least one worker agent; inputting the expanded query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to the user query; and providing the result data to the user account; wherein the step of generating an expanded query by performing query expansion on the user query through the at least one worker agent comprises: obtaining context data from pre-specified bookmark information in the user account or from the file system of a user terminal connected to the user account based on a specific tag command included in the user query; obtaining real-time user data including the current location and current time of the user account, and obtaining real-time situation data corresponding to the real-time user data from an external API; and generating the expanded query by combining the user query, the context data, the real-time user data, and the real-time situation data. A method for providing a creator support service based on a multi-agent artificial intelligence collaboration system, comprising: a step of generating an expanded query by performing query expansion on a user query through at least one worker agent, wherein if the length of the expanded query exceeds a threshold, a step of extracting core keywords and relationships from the expanded query; and a step of generating an optimal expanded query by summarizing and compressing the expanded query based on the core keywords and relationships. Claim 3 A method for providing a creator support service based on a multi-agent artificial intelligence collaboration system, performed by a computing device comprising at least one processor, comprising: obtaining a user query from a user account; determining at least one worker agent corresponding to the user query through a master agent; generating an expanded query by performing query expansion on the user query through the at least one worker agent; inputting the expanded query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to the user query; and providing the result data to the user account; wherein the step of inputting the expanded query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to the user query comprises: obtaining persona information and output format information set in the user account; generating a prompt including the expanded query, the persona information, and the output format information; and inputting the prompt into the large-scale language model to obtain result data reflecting the persona information and the output format information. Claim 4 A method performed by a computing device comprising at least one processor, comprising: obtaining a user query from a user account; determining at least one worker agent corresponding to the user query through a master agent; generating an expanded query by performing query expansion on the user query through the at least one worker agent; inputting the expanded query into a pre-trained large-scale language model (LLM) to obtain result data corresponding to the user query; providing the result data to the user account; crawling a plurality of source data from the web to extract knowledge data including entities and relationships between said entities; calculating weights for relationships between said entities based on a PageRank algorithm; constructing a knowledge base used to obtain the expanded query based on the knowledge data and said weights; periodically searching the web for new entities associated with entities included in said knowledge base; and, when said new entities are found, adding new knowledge data to said knowledge base including said new entities and relationships between said new entities and other entities previously stored. A method for providing creator support services based on a multi-agent artificial intelligence collaboration system, including Claim 5 A method for providing a creator support service based on a multi-agent artificial intelligence collaboration system, comprising: a step of obtaining a user query from a user account; a step of determining at least one worker agent corresponding to the user query through a master agent; a step of generating an expanded query by performing query expansion on the user query through the at least one worker agent; a step of obtaining result data corresponding to the user query by inputting the expanded query into a pre-trained large-scale language model (LLM); a step of providing the result data to the user account; a step of creating a post based on the result data, scheduling the publication of the post to a social media platform set in the user account, and providing an automatic response to comments on the post; a step of providing performance indicators for posts published to date through the user account; a step of searching a plurality of similar accounts that publish posts similar to the post from a knowledge base; and a step of generating guidance information based on the content publishing pattern and engagement characteristics of a specific similar account among the plurality of similar accounts that has obtained a higher performance indicator than the user account, and providing the guidance information to the user account. Claim 6 A method for providing a creator support service based on a multi-agent artificial intelligence collaboration system, wherein, in any one of claims 1 to 5, the step of determining at least one worker agent corresponding to the user query through the master agent comprises: a step of extracting at least one entity name from the user query; a step of extracting a user intent from the user query; a step of extracting the conversation context of the current conversation from a conversation history containing the user query; and a step of determining the at least one worker agent based on the at least one entity name, the user intent, and the conversation context. Claim 7 An apparatus comprising: a memory for storing one or more instructions; and a processor for executing the one or more instructions stored in the memory, wherein the processor performs the method of any one of claims 1 to 5 by executing the one or more instructions. Claim 8 A computer program stored on a computer-readable recording medium that is combined with a computer, which is hardware, to enable the method of any one of claims 1 to 5 to be performed. Claim 9 delete Claim 10 delete

Images