Method of automatically performing tasks determined based on user dialog input

Abstract

To provide a method and system for automatically performing tasks according to task context data required to perform tasks, as acquired based on the user dialog input.SOLUTION: A method of automatically performing a task determined according to a user dialog input using a computing system including a memory and processor is provided, the method comprising receiving a user dialog input, determining the type of a task requested by a user based on a result of analysis of the user dialog input, obtaining task context data required to perform the task based on the user dialog input, and performing the task according to the task context data.SELECTED DRAWING: Figure 13

Description

[Technical Field]

[0001] The present disclosure relates to a method for automatically executing a task determined in response to a user interactive input, and more particularly to a method and system for automatically executing a user-requested task determined in response to the user interactive input, in a systematic manner based on task context data acquired based on the user interactive input. [Background technology]

[0002] Recently, with the development of artificial intelligence (AI) technology, various services utilizing AI have been commercialized in various industrial fields. This AI technology uses an artificial neural network model to learn from a huge amount of data and output and provide the information desired by the user.

[0003] Meanwhile, conversational AI models are known that can interact with users, rather than simply learning data patterns and providing output values for input values. AI secretaries using such conversational AI models can be applied to smart devices to act as personal assistants, can be applied to chatbots that answer questions from corporate customers, and can be applied to smart home systems that can control the operation of home appliances connected to the Internet of Things according to user requests. As a result, conversational AI models can provide users with more convenient and user-friendly experiences in various technical fields.

[0004] In addition, active research is being conducted into artificial intelligence technology that can perform tasks requested by users, and for this reason, there have been many attempts to apply analysis methods based on task-oriented interaction graphs to interactive artificial intelligence models.

[0005] However, most research related to task-oriented dialogue graphs has been limited to either generating dialogue models using dialogue graphs drawn directly by humans or utilizing rule-based systems that automatically infer dialogue graphs based on previously learned dialogue policies. As a result, they are unable to provide dialogue graph-based dialogue models that are automatically constructed based solely on dialogue datasets without human intervention, and are unable to model diverse dialogue flows.

[0006] On the other hand, artificial intelligence is generally realized through a large number of AI models and deep learning based on them.

[0007] Such artificial intelligence is being developed to provide a variety of services taking into account the user's context (e.g., context, environment, and / or intent).

[0008] However, when processing a specific task based on a large amount of data, there is a limit in that the required computation cost and time are considerable.

[0009] This means that there are certain limitations to the use of AI models in on-device environments, which have recently attracted attention.

[0010] To solve this problem, model architectures such as MoE (Mixture of Experts) have traditionally been used.

[0011] Here, MoE refers to a machine learning model architecture that combines multiple expert models to solve complex problems.

[0012] Such an MoE may include expert models, which are multiple small networks designed to learn different portions and / or different characteristics of given data and perform data processing operations accordingly, and a gating network that evaluates the performance of each expert model and, based on this, determines which expert model is best suited to be assigned a particular task in response to the given data.

[0013] Therefore, according to the MoE architecture, a gating network receives given input data and determines a probabilistic or deterministic task assignment for each expert model, and the selected expert models then process the data for a specific task by performing their own tasks and returning the results.

[0014] By utilizing such MoE, an AI model can increase its overall efficiency and performance by activating only certain parts and concentrating computational resources when dealing with complex tasks or large data sets.

[0015] However, conventional MoE not only requires a high level of VRAM, but also presents a number of issues that must be resolved during the fine tuning process.

[0016] In addition, traditional MoE methods are designed to efficiently manage large-sized models, but they have limitations in supporting the efficiency of remaining resources that are not activated according to the given task.

[0017] Furthermore, in the conventional technology field, services are provided using AI models that are generally implemented for general purposes, but there is a problem in that it is difficult to quickly and easily ensure the AI analysis performance that is most suitable for a given context. Summary of the Invention [Problem to be solved by the invention]

[0018] Various embodiments of the present disclosure aim to provide a method and system for automatically executing a task, which determines the type of task requested by a user based on the analysis results of the user's interactive input, and systematically performs the task automatically based on task context data required to perform the task, obtained based on the user's interactive input.

[0019] However, the technical objectives to be achieved by the various embodiments of the present disclosure are not limited to the above-mentioned technical objectives, and other technical objectives may exist. [Means for solving the problem]

[0020] One embodiment provides a method for automatically executing a task determined in response to user input by a computing system including a memory and a processor, the method including the steps of receiving user interactive input, determining a type of task requested by a user based on an analysis result of the user interactive input, obtaining task context data required to perform the task based on the user interactive input, and performing the task based on the task context data.

[0021] In another aspect, the step of performing the task may include determining a task execution system required for executing the task based on task type information, and executing the task execution system based on the task context data to perform the task.

[0022] In another aspect, the method for automatically executing a task determined in response to user interactive input may further include receiving task-related information data related to the task, and performing the task may include performing the task based on at least one of the task context data and the task-related information data.

[0023] In another aspect, the task-related information data may include text data, and performing the task may include capturing the task-related information in the text format to obtain task-related information capture data, generating programming code for performing the task based on an analysis of the task-related information capture data, and executing the programming code to perform the task.

[0024] In another aspect, when the task execution system is executed to perform the task, it may automatically connect to a predetermined external system and complete the task execution by performing a predetermined operation on the predetermined external system.

[0025] In another aspect, the method for automatically executing a task determined in response to a user dialogue input further includes determining and providing a response dialogue act for the user input based on a dialogue graph, and determining a context of the goal-oriented dialogue by analyzing data of a series of goal-oriented dialogues including the user input and the response dialogue act, wherein the step of determining a type of task may determine the type of task requested by the user based on the context of the goal-oriented dialogue.

[0026] In another aspect, the step of determining the context of the goal-oriented dialogue may include extracting a plurality of keywords from the data of the goal-oriented dialogue, and analyzing at least one of the correlations between a plurality of dialogue acts included in the goal-oriented dialogue, the intention of the goal-oriented dialogue, and the purpose of the goal-oriented dialogue based on the plurality of keywords to determine the context.

[0027] In another aspect, the method for automatically executing a task determined in response to a user dialogue input may further include determining at least one task execution model from among a plurality of task execution models that is optimized for the task, the type of which is determined based on the context of the goal-oriented dialogue, and performing the task using the determined at least one task execution model.

[0028] In another aspect, the steps of determining the type of task and performing the task may involve the computing system determining the type of task and supporting predetermined operations required to perform the task.

[0029] In another aspect, the step of performing the task may include analyzing programming code related to the task, generating programming code for performing the task based on the analysis results, and executing the programming code to perform the task.

[0030] In another aspect, the method for automatically executing a task determined in response to user interaction input may further include a step of capturing a screen of an electronic device used by the user to obtain a user screen screenshot by receiving the user interaction input, and in the step of determining the type of task, the type of task may be determined based on information about the determined context of the goal-oriented interaction and an analysis of the user screen screenshot.

[0031] In another aspect, determining and providing a response dialogue act may include generating a dialogue graph modeling at least one conditional relationship for a dialogue dataset; sampling a plurality of dialogue act groups for responding to the user dialogue input using a previously trained dialogue model; adjusting the plurality of dialogue act groups based on the dialogue graph; and selecting one of the plurality of dialogue act groups that satisfies a predetermined condition.

[0032] In another aspect, the at least one conditional relationship may include at least one of a first conditional relationship regarding what utterance should be made for one utterance in a dialogue flow, a second conditional relationship regarding what utterance can be made for one utterance, and a third conditional relationship regarding what utterance cannot be made for one utterance.

[0033] In another aspect, the step of selecting one dialogue act group may include selecting one dialogue act group from the plurality of dialogue act groups that best satisfies the at least one conditional relationship.

[0034] One embodiment provides a system for automatically executing a task determined in response to a user interactive input, including at least one memory and at least one processor that reads at least one instruction word stored in the memory to perform a method for automatically executing a task determined in response to a user interactive input, wherein the at least one processor receives the user interactive input, determines a type of task requested by a user based on an analysis result of the user interactive input, obtains task context data required to perform the task from data related to the user interactive input, and performs the task based on the task context data.

[0035] One embodiment provides a system for automatically executing a task determined in response to a user interactive input, including an electronic device that receives user interactive input, and a computing device including at least one memory and at least one processor that reads at least one instruction word stored in the at least one memory to perform a method for automatically executing a task determined in response to the user interactive input, wherein the at least one processor receives the user interactive input, determines a type of task requested by a user based on an analysis result of the user interactive input, obtains task context data necessary to perform the task based on the user interactive input, and performs the task based on the task context data. [Effects of the Invention]

[0036] Various embodiments of the present disclosure may provide a task execution method and system that can automatically and accurately execute a user-requested task in a systematic manner by determining the type of task requested by a user based on analysis results of the user's interactive input, and obtaining basic information data required to perform the task from the user's interactive input, or obtaining information data related to keywords included in the user's interactive input from a separate database and utilizing the obtained data to execute the task whose type has been determined.

[0037] However, the effects obtained by the various embodiments of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood from the following description. [Brief explanation of the drawings]

[0038] [Figure 1] 1 illustrates an example block diagram of a computing system implementing goal-oriented interaction services and task execution services according to one embodiment. [Figure 2] 1 is a conceptual diagram illustrating how a computing system performs a user-requested task based on a goal-oriented interactive service according to one embodiment. [Figure 3] FIG. 1 is a simplified diagram illustrating the structure of neuromorphic circuitry that may be included in a processor according to one embodiment. [Figure 4] FIG. 1 is a block diagram of a computing device implementing goal-oriented interaction services and task execution services according to one embodiment. [Figure 5] FIG. 10 is a block diagram of a computing device implementing goal-oriented interaction services and task execution services according to another embodiment. [Figure 6] FIG. 10 is a block diagram of a computing device implementing a goal-oriented interaction service according to another embodiment. [Figure 7] 1 is a conceptual diagram illustrating a process for performing a goal-oriented interactive service according to an embodiment; [Figure 8]1 illustrates an internal block diagram of an AI agent model according to one embodiment. [Figure 9] 1 is a flowchart of a goal-directed interaction method according to one embodiment. [Figure 10] 1 is a conceptual diagram illustrating each step of a goal-directed interaction method according to one embodiment. [Figure 11] 10 is a table illustrating a performance evaluation result of a goal-oriented dialogue method according to an embodiment. [Figure 12] 1 is a flowchart of a method for context-based task execution of goal-directed dialogue through a dialogue model according to one embodiment. [Figure 13] 1 is a flowchart of a method for automatically executing a task determined in response to user interactive input according to one embodiment. [Figure 14] 14 is a flowchart illustrating steps for performing the tasks of the method of FIG. 13 in accordance with one embodiment. [Figure 15] 1 is a flowchart illustrating an MoE-based model identification method according to one embodiment. [Figure 16] FIG. 1 is a conceptual diagram illustrating an MoE-based model identification method according to one embodiment. [Figure 17] FIG. 10 illustrates an example of specialized model characteristic information according to an embodiment. [Figure 18] 1 is a flowchart illustrating a method for providing an AI agent based on an MoE application LLM according to an embodiment. [Figure 19] FIG. 1 is a conceptual diagram illustrating a method for providing an AI agent based on an MoE-based LLM according to one embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0039] Because the present invention can be modified in various ways and can have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. The advantages and features of the present invention, as well as methods for achieving them, will become clearer with reference to the following detailed embodiments, along with the drawings. However, the present invention is not limited to the embodiments disclosed below and can be realized in various forms. In the following embodiments, terms such as "first," "second," etc., are used to distinguish one element from another element without any limiting meaning. Furthermore, singular terms include plural terms unless the context clearly indicates otherwise. Furthermore, terms such as "include" or "have" mean the presence of a feature or element described in the specification and do not preclude the possibility that one or more other features or elements may be added. In the drawings, the size of elements may be exaggerated or reduced for ease of explanation. For example, the size and thickness of each element in the drawings are arbitrarily illustrated for ease of explanation, and the present invention is not necessarily limited to the contents shown.

[0040] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. When describing with reference to the drawings, the same or corresponding elements will be given the same reference numerals, and redundant description thereof will be omitted.

[0041] -System 1000 for providing purpose-oriented dialogue services

[0042] In one embodiment, the system 1000 generates a task-oriented dialogue graph based on an analysis of a dialogue dataset, and then selects a final response dialogue act by verifying and adjusting various response dialogue acts sampled by a previously trained dialogue model in response to a user dialogue input based on the generated dialogue graph, and provides the final response dialogue act as a response to the user dialogue input.

[0043] In this case, the system 1000 can provide the user with more reliable response dialogue acts by modeling predetermined conditional relationships on the dialogue dataset to generate a dialogue graph, and validating and adjusting the sampled response dialogue acts based on the predetermined conditional relationships included in the dialogue graph.

[0044] Furthermore, the system 1000 can provide the user with a more convenient task processing experience by determining the type of task requested by the user based on the user interaction input and the ultimately generated response interaction act, and efficiently performing the task based on the determined task type.

[0045] FIG. 1 illustrates an example block diagram of a computing system 1000 that implements goal-oriented interaction and task execution services according to one embodiment.

[0046] As shown in FIG. 1 , a computing system 1000 implementing goal-oriented interaction services and task execution services according to one embodiment includes a user computing device 110, a server computing system 130, and a training computing system 150, where the devices are capable of communicating via a network 170.

[0047] The goal-oriented interaction method according to one embodiment may be 1) implemented and provided locally by the user computing device 110, 2) implemented and provided in the form of a web service by the server computing system 130 that communicates with the user computing device 110, or 3) implemented and provided by the user computing device 110 and the server computing system 130 in cooperation with each other.

[0048] Here, in an embodiment, the user computing device 110 and / or the server computing system 130 may train the machine learning models 120 and / or 140 by interacting with a training computing system 150 communicatively coupled via a network 170. The training computing system 150 may be separate from the server computing system 130 or may be part of the server computing system 130.

[0049] Here, the AI model can be 1) trained directly locally by the user computing device 110, 2) trained by the server computing system 130 and the user computing device 110 interacting with each other via the network 170, or 3) trained using various training and learning techniques by a separate training computing system 150. The AI model trained by the training computing system 150 can also be transmitted to the user computing device 110 and / or the server computing system 130 via the network 170 and provided / updated.

[0050] In some embodiments, the training computing system 150 may be part of the server computing system 130 or part of the user computing device 110 .

[0051] FIG. 2 is a conceptual diagram illustrating how a computing system 1000 according to one embodiment performs a user-requested task based on a goal-oriented interactive service.

[0052] 2, the computing system 1000 can receive various forms of user interaction input, provide appropriate responsive interaction acts to the received user interaction input, and perform tasks requested by a user based on the user interaction input and the responsive interaction acts, where the tasks may include various types of tasks determined based on goal-directed interactions including the user interaction input and the responsive interaction acts.

[0053] The computing system 1000 understands the context of a series of goal-directed interactions consisting of user interaction inputs and corresponding response interactions.

[0054] The computing system 1000 can analyze goal-oriented dialogue patterns consisting of various types of user dialogue inputs and corresponding response dialogue actions, and based on this, determine context related to the intention, purpose, etc. of the corresponding dialogue.

[0055] The computing system 1000 can extract multiple keywords contained in the data of the goal-oriented dialogue, and based on the extracted multiple keywords, analyze the correlations between multiple dialogue acts contained in the dialogue, the intention and purpose of the dialogue, etc., to ultimately determine the context of the goal-oriented dialogue.

[0056] Additionally, the computing system 1000 can determine the type of task requested by the user based on information about the context of the goal-oriented interaction.

[0057] For example, multiple keywords (e.g., hotel, reservation, date, number of guests, 5-star, etc.) are extracted from a series of dialogue data including a user dialogue input requesting a hotel reservation and a response dialogue act requesting information about the hotel reservation (e.g., reservation date, number of guests, hotel grade, etc.).

[0058] In addition, the context of the corresponding dialogue can be determined by analyzing at least one of the correlation between the user dialogue action included in the corresponding dialogue and the response dialogue action provided by the computing system 1000, the intention and purpose of the corresponding dialogue, based on the extracted keywords.

[0059] For example, according to the keyword-based context determination process, a dialogue context containing multiple dialogue acts such as "Please make a hotel reservation", "When is the check-in date?", "Check-in on December 31, 2024 and check-out on January 3, 2025", "How many people will be staying?", "3 people" is determined to indicate that a user is requesting a hotel reservation and thus an automated hotel reservation service is needed.

[0060] Furthermore, the type of task requested by the user can be determined based on the determined interaction context.

[0061] For example, based on the context of the interaction in which it is determined that an automatic hotel reservation service is necessary in response to a user's hotel reservation request, the task requested by the user is determined to be "book a hotel."

[0062] Meanwhile, the number of task types determined based on the context of the goal-oriented dialogue may be multiple. The goal-oriented dialogue may include various types of user dialogue inputs and various response dialogue actions thereto, and the context of such goal-oriented dialogue may be associated with various tasks. Thus, according to an embodiment, multiple task types may be determined based on the context of the goal-oriented dialogue.

[0063] The system 1000 executes a task whose type is determined based on a goal-directed dialogue including a user dialogue input and a response dialogue act, and provides the execution result to the user.

[0064] For example, if the task type is determined to be "hotel reservation," the system 1000 can complete a hotel reservation task to meet the user's request based on data associated with a series of goal-directed interactions.

[0065] In this case, the system 1000 determines the type of task and supports the predetermined operations required to perform the typed task.

[0066] For example, the operating system of the system 1000 can control the processors 111, 131 to directly determine the type of task and perform the predetermined operations required to perform the determined type of task.

[0067] Also, for example, the operating system of system 1000 may determine the type of task and provide a predetermined API (Application programming interface) and / or SDK (Software development kit) to support the predetermined operations required to perform the determined type of task.

[0068] The system 1000 generates and executes programming code necessary to perform a task whose type is determined based on the context of the goal-oriented dialogue, where the programming code may be code written to perform a particular task using a programming language (e.g., Java, Python, JavaScript, etc.).

[0069] Meanwhile, the system 1000 can receive user interactive input to capture the screen of the electronic device used by the user to obtain a user screen screenshot.

[0070] Thereafter, the system 1000 may determine the type of task based on information about the determined goal-oriented interaction context and an analysis of the user screen shot.

[0071] For example, if the dialogue context is determined to be "hotel reservation" and the user screen screenshot includes a screen of a homepage that provides hotel reservation services, the system 1000 can determine the task type as "book a hotel through the homepage that provides the corresponding hotel reservation service."

[0072] In this case, the system 1000 can automatically perform a series of actions (for example, cursor movement, clicks, text input, etc.) required to perform the determined task on the user screen.

[0073] However, without being limited thereto, the task determined based on the goal-directed dialogue may be determined into an infinite number of types depending on the content of the user dialogue input and the response dialogue act, for example, the task may be determined into various types such as purchasing a product, sending an email, searching for information, creating a document, etc.

[0074] Furthermore, in one embodiment, when the type of a task is determined based on the context of the goal-oriented dialogue, the system 1000 determines at least one task execution model from among a plurality of task execution models that is optimized for the task whose type is determined based on the context of the goal-oriented dialogue, and performs the task using the determined at least one task execution model.

[0075] In this case, the method by which the system 1000 determines at least one task execution model optimized for the task and uses it to perform the task is substantially the same as the "MoE-based model identification method" described below, and therefore a detailed description thereof will be omitted.

[0076] Also, for example, system 1000 may include a first user computing device 180 of a first type, a second user computing device 181 of a second type, a third user computing device 182 of a third type, and a fourth user computing device 183 of a fourth type that can receive various forms of user interaction input from users.

[0077] Here, the user interaction input may have at least one form of text, voice, gesture, and touch, but is not limited thereto, and the user interaction input may have various forms other than the above examples.

[0078] The first user computing device 180 of the first type may be a virtual reality electronic device, the second user computing device 181 of the second type may be a mobile electronic device, the third user computing device 182 of the third type may be an augmented reality electronic device, and the fourth user computing device 183 of the fourth type may be a desktop.

[0079] However, without limitation, system 1000 may include various forms of user computing devices other than those illustrated above that are capable of receiving user interactive input.

[0080] -User Computing Device (110: User Computing Device)

[0081] The user computing devices 110 may include any other type of computing device, such as smartphones, mobile phones, digital broadcasting devices, personal digital assistants (PDAs), portable multimedia players (PMPs), desktops, wearable devices, embedded computing devices, tablet PCs, augmented reality (VR) devices, and / or virtual reality (AR) devices.

[0082] Such a user computing device 110 may include at least one or more processors 111 and memory 112, where the processor 111 may comprise at least one or a plurality of electrically connected processors including central processing units (CPUs), graphics processing units (GPUs), application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0083] For example, ASICs may have the structure of an array of neuromorphic circuits that include multiple neuron circuits.

[0084] As shown in FIG. 3, for example, the neuromorphic circuit 300 may include a plurality of presynaptic neuron circuits 310, a plurality of presynaptic lines 311 extending horizontally from the plurality of presynaptic neuron circuits 310, a plurality of postsynaptic neuron circuits 320, a plurality of postsynaptic lines 321 extending vertically from the plurality of postsynaptic neuron circuits 320, and synaptic circuits 330 provided at the intersections of the plurality of presynaptic lines 311 and the plurality of postsynaptic lines 321.

[0085] The plurality of presynaptic neuron circuits 310 can transmit signals input from the outside to the plurality of synapse circuits 330 via the plurality of presynaptic lines 311 in the form of electrical signals.

[0086] Additionally, the plurality of post-synaptic neuron circuits 320 can receive electrical signals from the plurality of synaptic circuits 330 via the plurality of post-synaptic lines 321 .

[0087] Additionally, the plurality of post-synaptic neuron circuits 320 may transmit electrical signals to the plurality of synaptic circuits 330 via the plurality of post-synaptic lines 321 .

[0088] The plurality of synapse circuits 330 can store weights included in layers that make up the neural network system realized by the neuromorphic circuit 300, and perform predetermined calculations based on the weights and input data.

[0089] For example, each of the plurality of synapse circuits 330 may include a resistive memory cell having a variable resistance, in which case the resistance value of the plurality of synapse circuits 330 changes depending on the voltage applied via the plurality of presynaptic neuron circuits 310 or the plurality of postsynaptic neuron circuits 320, and weighted value data corresponding to such resistance change can be stored.

[0090] The neuromorphic circuit 300 is formed by replicating the neuron and synapse structure, which are essential elements of the human brain. When a deep neural network (DNN) is realized using the neuromorphic circuit 300, it is possible to improve data processing speed and reduce power consumption compared to when using the existing von Neumann structure.

[0091] The memory 112 may include one or more non-transitory / transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., or a combination thereof, and may also include web storage of a server that performs memory storage functions on the Internet. Such memory 112 may store data 113 and instructions 114 required for the at least one processor 111 to perform functional operations, such as training an AI model or performing vision inspection through an AI model.

[0092] In one embodiment, the user computing device 110 may store at least one or more machine learning models 120.

[0093] For example, the machine learning model 120 may be a variety of machine learning models, such as multiple neural networks (e.g., deep neural networks) for performing goal-directed interaction methods and task execution methods, or other types of machine learning models including nonlinear and / or linear models, or may be composed of combinations thereof.

[0094] For example, the machine learning model may include linear regression, decision trees, random forests, gradient boosting, pre-trained language models, and / or deep learning models, etc. The neural network may include at least one of feed-forward neural networks, recurrent neural networks (e.g., long short-term memory recurrent neural networks), convolutional neural networks, and / or other forms of neural networks.

[0095] In addition, in various embodiments, the user computing device 110 can store models to be used in each step and prompt templates that serve as the basis for input to the model in order to execute at least some of the steps for the goal-oriented dialogue method and task execution method through a large-scale language model (LLM).

[0096] In one embodiment, the user computing device 110 receives at least one or more machine learning models 120 from the server computing system 130 via the network 170, stores them in the memory 112, and then executes the stored machine learning models 120 via the processor 111 to perform interaction dataset analysis, etc.

[0097] In another embodiment, the server computing system 130 may include at least one machine learning model 140, execute operations through the machine learning model 140, and interface with the user computing device 110 in a manner that communicates data associated therewith to provide goal-oriented interaction services and task execution services to the user.

[0098] For example, the user computing device 110 can perform goal-directed interactive services over the web in a manner in which the server computing system 130 utilizes the machine learning model 140 to provide output in response to user input.

[0099] The artificial intelligence models may also be implemented in such a way that at least some of the machine learning models 120 and / or 140 execute on the user computing device 110 and the remainder execute on the server computing system 130 .

[0100] The user computing device 110 may also include at least one or more input components 121 that sense user input. For example, the user input component 121 may include a touch sensor (e.g., a touch screen and / or a touch pad) that senses touch of a user's input medium (e.g., a finger or a stylus), an image sensor that senses user motion input, a microphone that senses user voice input, a button, a mouse and / or a keyboard, etc. The user input component 121 may also include an interface and an external controller (e.g., a mouse and / or a keyboard) when receiving input to the external controller via an interface.

[0101] -Server Computing System (130:Server Computing System)

[0102] The server computing system 130 performs a series of processes to provide goal-oriented interactive services.

[0103] The server computing system 130 may also perform a series of processes to provide a task execution service based on the context of goal-oriented dialogue.

[0104] In particular, in an embodiment, the server computing system 130 can provide goal-oriented interaction services by exchanging data with an external device, such as the user computing device 110, necessary to drive goal-oriented interaction service and task execution service processes in the external device.

[0105] More particularly, in an embodiment, the server computing system 130 may provide an environment in which applications can operate to provide goal-oriented interaction services and task execution services on the user computing device 110 .

[0106] To this end, the server computing system 130 may include application programs, data, and / or commands for applications to operate, and may transmit and receive various data based thereon to and from the external device.

[0107] The server computing system 130 may include at least one or more processors 131 and memory 132. Here, the processor 131 may be comprised of at least one or a plurality of electrically connected processors including central processing units (CPUs), graphics processing units (GPUs), application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0108] For example, ASICs may have the structure of an array of neuromorphic circuits that include multiple neuron circuits (see FIG. 3).

[0109] The memory 132 may include one or more non-transitory / transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and a combination thereof. Such memory 132 can store data 133 and instructions 134 required for the processor 131 to perform functional operations, such as training an artificial intelligence model or performing a goal-oriented interaction method and a task execution method through the artificial intelligence model.

[0110] In one embodiment, the server computing system 130 may be implemented with at least one or more computing devices. For example, the server computing system 130 may be implemented with multiple computing devices operating according to a sequential computing architecture, a parallel computing architecture, or a combination thereof. The server computing system 130 may also include multiple computing devices connected via a network 170.

[0111] The server computing system 130 may also store at least one or more machine learning models 140. For example, the server computing system 130 may include neural networks and / or other multi-layer nonlinear models as the machine learning models 140. Illustratively, the neural networks may include feedforward neural networks, deep neural networks, recurrent neural networks, and convolutional neural networks.

[0112] In an embodiment, the server computing system 130 may further include a data store computing system (hereinafter referred to as a data store), which is a storage for continuously storing and managing the original data underlying the goal-oriented interactive service.

[0113] Such data stores may include various forms of data storage, ranging from file systems to cloud storage. For example, the data stores may include at least one of the following databases: a relational database that uses Structured Query Language (SQL) to define and manipulate data; a NoSQL database that is designed for flexibility and scalability and handles unstructured and semi-structured data; a data warehouse, which is a system used for reporting and data analysis that centralizes large amounts of data from multiple sources and optimizes it for query and analysis; a data warehouse that stores large amounts of raw data in basic formats: structured, semi-structured, and unstructured; and a local storage device and / or network attached storage (NAS), which stores data in files in a format that is typically accessible by a computer operating system.

[0114] -Training Computing System (150:Training Computing System)

[0115] The training computing system 150 may include at least one or more processors 151 and memory 152. Here, the processor 151 may comprise at least one or a plurality of electrically connected processors including central processing units (CPUs), graphics processing units (GPUs), application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and / or other electrical units for performing functions.

[0116] For example, ASICs may have the structure of an array of neuromorphic circuits that include multiple neuron circuits (see FIG. 3).

[0117] The memory 152 may include one or more non-transitory / transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and a combination thereof. Such memory 152 can store data 153 and instructions 154 required for the processor 151 to perform tasks such as learning an artificial intelligence model.

[0118] For example, the training computing system 150 may include a model trainer 160 that uses various training or learning techniques, such as backpropagation of errors, to train the machine learning models 120 and / or 140 stored on the user computing device 110 and / or the server computing system 130.

[0119] Illustratively, such a model trainer 160 may perform backpropagation updates to one or more parameters of the machine learning models 120 and / or 140 for goal-oriented interactive services based on a defined loss function.

[0120] In some implementations, performing backpropagation of errors may include performing truncated backpropagation through time. The model trainer 160 can perform a number of generalization techniques (e.g., deweighting, dropout, and / or knowledge distillation, etc.) to improve the generalization ability of the trained machine learning models 120 and / or 140.

[0121] For example, model trainer 160 may train machine learning models 120 and / or 140 based on a set of training data 161, where training data 161 may include data of different modalities, such as, for example, images, audio samples, and / or text.

[0122] Training data 161 may also include, for example, various types of dialogue data, where the dialogue data may be data related to task-oriented dialogue requesting a particular task and providing a response to the requested task.

[0123] Examples of image types that can be used may include video frames, LiDAR point clouds, X-ray images, computed tomography scans, hyperspectral images, and / or various other forms of imagery.

[0124] Such training data 161 can be provided by the user computing device 110 and / or the server computing system 130. When the training computing device trains the machine learning models 120 and / or 140 on the specific data of the user computing device 110, the machine learning models 120 and / or 140 can be characterized as personalized models.

[0125] The model trainer 160 then contains the computer logic that is utilized to provide the desired functionality.

[0126] Additionally, model trainer 160 can be implemented in hardware, firmware, and / or software controlling a general-purpose processor. In one implementation, model trainer 160 includes a program file stored on a storage device, can be loaded into memory 152, and executed by one or more processors 151. In another implementation, model trainer 160 includes one or more sets of computer-executable data 153 and instructions 154 stored on a computer-readable storage medium of some type, such as RAM, a hard disk, or optical or magnetic media.

[0127] Network 170 may include, but is not limited to, a 3GPP (registered trademark) (3rd Generation Partnership Project) network, a LTE (Long Term Evolution) network, a WIMAX (World Interoperability for Microwave Access) network, the Internet, a LAN (Local Area Network), a Wireless LAN (Wireless Local Area Network), a WAN (Wide Area Network), a PAN (Personal Area Network), a Bluetooth (registered trademark) network, a satellite broadcast network, an analog broadcast network, and / or a DMB (Digital Multimedia Broadcasting) network.

[0128] In general, communications over network 170 may occur using any type of wired and / or wireless connection, and via a variety of communication protocols (e.g., TCP / IP, HTTP, SMTP and / or FTP, etc.), encodings or formats (e.g., HTML and / or XML, etc.), and / or protection schemes (e.g., VPN, Secure HTTP and / or SSL, etc.).

[0129] FIG. 4 is a block diagram of a computing device 100 implementing goal-oriented interaction and task execution services according to one embodiment.

[0130] 4, the computing devices 100 included in the user computing device 110, the server computing system 130, and the training computing system 150 include multiple applications (e.g., Application 1 through Application N). Each application may include a machine learning library and one or more machine learning models. For example, the applications may include an image processing (e.g., detection, classification, and / or segmentation) application, a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, and / or a chat-bot application.

[0131] In an embodiment, the computing device 100 may include a model trainer 160 for training an artificial intelligence model, and the trained artificial intelligence model may be stored and operated to provide output data in response to predetermined input data (e.g., an interaction data set, in an embodiment).

[0132] Each application on computing device 100 may communicate with numerous other components on computing device 100, such as, for example, at least one or more sensors, context managers, device state components, and / or additional components. In one embodiment, each application may communicate with each device component using an API (e.g., a public API). In one embodiment, the API used by each application may be specific to that application.

[0133] FIG. 5 is a block diagram of a computing device 200 implementing goal-oriented interaction services and task execution services according to another embodiment.

[0134] 5, computing device 200 includes multiple applications (e.g., Application 1 through Application N). Each application can communicate with a central intelligence layer. For example, the applications may include an image processing application, a text messaging application, an email application, a dictation application, a virtual keyboard application, and / or a browser application. In one embodiment, each application can communicate with the central intelligence layer (and the models stored therein) using an API (e.g., a common API across all applications).

[0135] The central intelligence layer may include multiple machine learning models. For example, as shown in FIG. 5, at least a portion of each machine learning model may be provided to each application and managed by the central intelligence layer. In other implementations, two or more applications may share a single machine learning model. For example, in some implementations, the central intelligence layer may provide a single model for all applications. In some implementations, the central intelligence layer may be included within or implemented separately from the operating system of computing device 200.

[0136] The central intelligence layer can communicate with a central device data layer, which can be a centralized data storage for the computing device 200. As shown in Figure 5, the central device data layer can communicate with numerous other components of the computing device 200, such as, for example, one or more sensors, a context manager, a device state component, and / or additional components. In some implementations, the central device data layer can communicate with each device component using an API (e.g., a private API).

[0137] The techniques described herein may refer to servers, databases, software applications, and other computer-based systems, as well as actions taken and information sent to or from such systems. It will be appreciated that the inherent flexibility of computer-based systems allows for a wide range of possible configurations, combinations, and divisions of work and functionality between and among components. For example, the processes described herein may be implemented using a single device or component or multiple devices or components operating in combination. Databases and applications may be implemented in a single system or in systems distributed across multiple systems. Distributed components may operate sequentially or in parallel.

[0138] FIG. 6 is a block diagram illustrating the functionality of a computing device 400 that implements a goal-oriented interaction service according to one embodiment.

[0139] As shown in FIG. 6, the computing device 400 may include a dialogue graph generation module 10, a dialogue act group sampling module 20, a dialogue act group adjustment module 30, a dialogue act group selection module 40, and a dialogue act selection module 50.

[0140] The dialogue graph generation module 10 can generate a dialogue graph based on an externally received dialogue dataset.

[0141] The dialogue dataset may include data relating to dialogues between a variety of speakers taking place in a variety of environments, for example, a dialogue dataset may include data relating to a variety of types of dialogue sequences, such as a dialogue between a customer and a reservation agent in the context of booking a hotel, a dialogue between a buyer and a seller in the context of purchasing an item, a dialogue between a tutor and a tutee interacting about an academic assignment, etc.

[0142] A dialogue graph is a graphical representation of the relationships between various types of utterances contained in a dialogue dataset, and may include multiple nodes corresponding to multiple dialogue acts that indicate the function of a specific utterance, and multiple edges that indicate various information regarding the relationships between multiple dialogue acts.

[0143] Here, for example, a dialogue act may include various types of speech functions such as a question, information, a request, and a confirmation.

[0144] Furthermore, the relationship between multiple dialogue acts may include a predetermined conditional relationship between two utterances. For example, the predetermined conditional relationship may include at least one of a first conditional relationship (Should relationship) regarding what utterance should be made for one utterance, a second conditional relationship (Can relationship) regarding what utterance can be made for one utterance in the dialogue flow, and a third conditional relationship (Should-not relationship) regarding what utterance must not be made for one utterance.

[0145] For example, as shown in FIG. 7, the dialogue graph generation module 10 generates a task-oriented dialogue flow (TOD-Flow) graph including a plurality of nodes including A, B, C, and D and a plurality of edges corresponding to the relationships between the plurality of nodes (S10).

[0146] The dialogue graph generation module 10 can vectorize the dialogue dataset by embedding the words and sentences included in the dialogue dataset. The dialogue graph generation module 10 can analyze the vectorized dialogue dataset to learn related information of the dialogue dataset, such as the intention of the dialogue utterances, dialogue acts, slots, values, context information, information on the relationship between utterances, and patterns of sequential dialogue flow.

[0147] For example, the interaction graph generation module 10 may use an artificial intelligence model based on at least one of a Transformer-based model, a recurrent neural network (RNN), and a long-term memory (LSTM) to learn relevant information from the interaction dataset.

[0148] Furthermore, the interaction graph generation module 10 can generate an interaction graph by modeling at least one conditional relationship for the interaction dataset based on the related information of the learned interaction dataset.

[0149] In this case, the interaction graph generation module 10 models at least one relationship for the interaction dataset by maximizing a loss function optimized for at least one conditional relationship.

[0150] For example, the dialogue graph generation module 10 generates a dialogue graph using a dialogue graph generation model that has been trained to maximize the expected value when the nth dialogue act satisfies the first conditional relation (Should relation) in the dialogue context.

[0151] In detail, the dialogue graph generation module 10 can generate a dialogue graph by utilizing a dialogue graph generation model that has been trained to maximize the expected value when the nth dialogue act satisfies the first conditional relationship (Should relationship) in a situation where the first conditional relationship (Should relationship) of the dialogue context must be satisfied.

[0152] For example, the dialogue graph generation module 10 can model a first conditional relation (Should relation) regarding what utterance should be made for one utterance in the dialogue flow by maximizing the loss function defined by the following [Equation 1].

[0153]

number

[0154] JPEG2025119575000003.jpg51151

[0155] Furthermore, for example, the dialogue graph generation module 10 can generate a dialogue graph by utilizing a dialogue graph generation model that has been trained to maximize the expected value when the nth dialogue act in the dialogue context satisfies the second conditional relationship (Can relationship) and at the same time does not satisfy the third conditional relationship (Should-not relationship).

[0156] In detail, the dialogue graph generation module 10 generates the dialogue graph using a dialogue graph generation model trained to maximize the sum of the first expectation when, in a situation where the nth dialogue act has occurred, the nth dialogue act in the dialogue context satisfies the second conditional relation (Can relation) but does not satisfy the third conditional relation (Should-not relation), and the second expectation when, in a situation where the nth dialogue act has not occurred, the nth dialogue act in the dialogue context does not satisfy the second conditional relation (Can relation) but satisfies the third conditional relation (Should-not relation). For example, the dialogue graph generation module 10 can model the second conditional relation (Can relation) regarding what utterances can be made for one utterance in the dialogue flow, and the third conditional relation (Should-not) regarding what utterances must not be made for one utterance, by maximizing the loss function defined by the following [Equation 2]:

[0157]

number

[0158] JPEG2025119575000005.jpg58151

[0159] The dialogue act group sampling module 20 can sample a plurality of dialogue act groups for responding to an externally received user dialogue input.

[0160] For example, the dialogue act group sampling module 20 can sample a plurality of dialogue act groups appropriate as responses to a user dialogue input received via an electronic device. As shown in Fig. 7, a next dialogue act prediction can be performed for the user dialogue input. For example, for a user dialogue input related to "book a hotel," the dialogue act group sampling module 20 samples dialogue act groups including various types of dialogue acts such as "confirm information" and "book hotel" as dialogue act groups that are responses following the user dialogue input (S20).

[0161] The dialogue act group sampling module 20 may include an artificial neural network structure that can extract features from various types of dialogue datasets and already be trained to provide appropriate output data for input data. For example, the dialogue act group sampling module 20 may include a Transformer-based neural network architecture (e.g., GPT-3, GPT-4, BERT-based models, etc.).

[0162] The dialogue act group adjustment module 30 can adjust the sampled dialogue act groups based on the generated dialogue graph.

[0163] Each of the multiple dialogue act groups generated by the dialogue act group sampling module 20 may include at least one dialogue act associated with the user dialogue input.

[0164] The dialogue act group adjustment module 30 can adjust the plurality of dialogue act groups based on whether each of the plurality of dialogue act groups generated by the dialogue act group sampling module 20 in response to the user dialogue input satisfies at least one condition relationship included in the dialogue graph.

[0165] For example, the dialogue act group adjustment module 30 can determine whether each of a plurality of dialogue act groups satisfies a first condition relationship (Should relationship) with respect to the user dialogue input, and add a dialogue act that satisfies the first condition relationship (Should relationship) with respect to the user dialogue input to a dialogue act group that does not satisfy the first condition relationship (Should relationship).

[0166] In addition, the dialogue act group adjustment module 30 can determine whether each of the multiple dialogue act groups satisfies the second condition relationship (Can relationship) and the anti-third condition relationship (not should not) with respect to the user dialogue input, and remove a dialogue act that does not satisfy the second condition relationship (Can relationship) and the anti-third condition relationship (not should not) from at least one dialogue act included in a dialogue act group that does not satisfy the second condition relationship (Can relationship) and the anti-third condition relationship (not should not).

[0167] In this way, the dialogue act group adjustment module 30 adjusts the sampled dialogue act groups to match the dialogue graph (S20), thereby improving the reliability of the goal-oriented dialogue service provided by the system 1000 and the controllability of the dialogue model.

[0168] The dialogue act group selection module 40 can select any one of the adjusted dialogue act groups that satisfies a predetermined condition.

[0169] For example, the dialogue act group selection module 40 ranks the adjusted dialogue act groups in descending order of the number of dialogue acts that satisfy at least one conditional relationship in the dialogue graph (S30).

[0170] For example, a first dialogue act group, a second dialogue act group, and a third dialogue act group can be adjusted by the dialogue act group adjustment module 30, and after the adjustment work is completed, the first dialogue act group contains three dialogue acts, the second dialogue act group contains two dialogue acts, and the third dialogue act group contains four dialogue acts.

[0171] In this case, the dialogue act group selection module 40 can select the third dialogue act group as the first priority, which contains the most dialogue acts after being adjusted based on at least one conditional relationship in the dialogue graph, from the first to third dialogue act groups.

[0172] The dialogue act selection module 50 can select any one of at least one dialogue act included in one dialogue act group selected from a plurality of dialogue act groups to provide a response output to a user dialogue input.

[0173] For example, the dialogue act selection module 50 may calculate the relevance of at least one dialogue act included in the selected dialogue act group to the user's dialogue input using a probability. Thereafter, the dialogue act selection module 50 may select one dialogue act from the selected dialogue act group that has the highest calculated probability of relevance to the user's dialogue input, and provide the selected dialogue act as a response to the user.

[0174] -AIAM: Artificial Intelligence Agent Specialization Model

[0175] In another aspect, the computing system 1000 as described above may include an AI Agent Specialization Model (AIAM) according to one embodiment.

[0176] Here, the AI agent specialized model (AIAM) according to an embodiment is an AI agent model that applies the MoE (Mixture of Experts) architecture realized according to an embodiment, and may be an artificial intelligence model including data processing algorithms that can act autonomously in a specific environment, solve tasks, and achieve goals. Here, MoE refers to a machine learning model architecture that combines multiple expert models to solve complex problems.

[0177] Such an AI agent specialization model (AIAM) may include data processing algorithms to realize cognitive ability to collect and interpret data from a given environment, decision mechanism to determine optimal actions based on the collected data, execution ability to carry out the determined actions, and learning ability to improve behavior through experience.

[0178] In one embodiment, the AI agent specialized model (AIAM) can acquire predetermined input data (e.g., text, voice, image, video, and / or sensing data based on a specific sensor, etc.) and provide output data (e.g., response data to a specific question and / or control signals based on a specific command word, etc.) by performing a predetermined task based on the acquired input data.

[0179] FIG. 8 shows an internal block diagram of an AI agent model according to one embodiment.

[0180] In particular, as shown in FIG. 8, an AI agent model according to one embodiment may include at least one router (RT: Router, Gating Network), orchestrator (OCT: Orchestrator), sLLM (small Large Language Model), general MoE model (NM: Normal MoE Model), external model (EM: External Model), and / or specialized model (SM: Specialized Model).

[0181] Here, in FIG. 8, to avoid obscuring the features of various embodiments, the AI agent model is described as including the above-described components.

[0182] However, it is obvious that a person skilled in the art would understand that, depending on the embodiment, other general-purpose components may be included in addition to the components shown in FIG. 8, or some of the components shown in FIG. 8 may be omitted.

[0183] More specifically, a router (RT: Router, Gating Network) according to one embodiment may be an artificial intelligence module that performs task allocation and / or traffic adjustment for multiple models in the MoE architecture.

[0184] Specifically, the router (RT) can analyze given input data and / or requested tasks, etc., and determine which model is most suitable for processing that data.

[0185] In this case, in an embodiment, the router (RT) can determine a model that is optimized for given data processing based on the performance, expertise, and / or previous experience of each model.

[0186] In addition, the router (RT) can support efficient data processing by distributing a given task to at least one model, taking into account the system load.

[0187] Furthermore, the router (RT) can adjust the tasks assigned to a particular model in response to changes in the real-time system.

[0188] In an embodiment, such a router (RT) may be an artificial intelligence module that selectively determines a model (hereinafter, a domain-specific model) that performs data processing operations optimized for a given domain.

[0189] That is, in an embodiment, the router (RT) may be an artificial intelligence module that selects a model (i.e., a domain-specific model) from among multiple models included in the AI agent specialization model (AIAM) that is determined to perform data processing (such as deep learning in an embodiment) that is most suitable for a given domain.

[0190] For reference, a domain in accordance with an embodiment refers to the data, rules, terminology, problem definition, and / or processes, etc., that a given AI system uses to perform a particular task.

[0191] In an embodiment, the router (RT) performs data analysis based on the characteristics of predetermined input data (e.g., user input and / or specific sensing data, etc.) and / or the required task, and based on this, grasps the data processing characteristics optimized for the corresponding task, detects a predetermined model that realizes this, and determines a domain-specific model.

[0192] In other words, the router (RT) according to the embodiment may be an artificial intelligence module that detects the model that can most effectively perform data processing for a given domain, allocates / distributes the corresponding task processing work, and manages it.

[0193] Here, the router (RT) according to the embodiment may include a router (RT) that has already been learned according to the disclosed predetermined algorithm, a router (RT) that has been additionally learned according to one embodiment, and / or a router (RT) that has been newly learned according to a new method.

[0194] Meanwhile, an orchestrator (OCT) according to one embodiment may be an artificial intelligence module that controls and manages the overall configuration of the AI agent specialization model (AIAM).

[0195] In particular, in an embodiment, the orchestrator (OCT) allocates various tasks generated from the overall system to appropriate resources (in an embodiment, routers (RT) and / or predetermined models, etc.).

[0196] The orchestrator (OCT) also manages the available models and resources such as hardware resources (e.g., CPU and / or GPU) to ensure efficient use.

[0197] In addition, the orchestrator (OCT) monitors the performance of the entire system and adjusts specific parameters as needed or optimizes the network configuration.

[0198] The orchestrator (OCT) manages the interaction between multiple routers (RT) and / or models, and controls the data flow and processing process.

[0199] That is, in the embodiment, the orchestrator (OCT) controls and manages the entire system of the AI agent specialization model (AIAM) and performs the role of a main router (RT) that controls at least one router (RT).

[0200] Here, the orchestrator (OCT) and the router (RT) according to the embodiment can closely cooperate with each other to support efficient operation of the MoE system.

[0201] Specifically, the orchestration (OCT) monitors the performance of the router (RT) as the administrator of the entire system and adjusts the router (RT) strategy as necessary.

[0202] On the other hand, the router (RT) can essentially allocate data processing tasks according to the instructions of the orchestrator (OCT) and / or its own algorithms to achieve efficient system control.

[0203] In an embodiment, the Orchestrator (OCT) and / or Router (RT) as described above may be a Master Model (P) that can control and manage the overall system and / or the remaining configuration of the AI agent model (i.e., sLLM, general MoE model (NM), external model (EM) and / or specialized model (SM), etc.).

[0204] Meanwhile, according to one embodiment, a small Large Language Model (sLLM) is an artificial intelligence module realized as a lightweight version of a Large Language Model (LLM).

[0205] In other words, sLLM is an artificial intelligence module built to achieve similar performance to large models such as LLMs with fewer resources.

[0206] In an embodiment, such an sLLM may include a plurality of specialist models (SMs) and a router (RT) coupling-based MoE model (in an embodiment, an MoELM) according to an embodiment. Additionally, the sLLM may include a domain-specific specialist model-based MoE model (in an embodiment, a DMoE model) according to an embodiment.

[0207] Furthermore, a normal MoE model (NM) according to an embodiment of the present invention may refer to a predetermined MoE model realized by the disclosed universal method.

[0208] Illustratively, the general MoE model (NM) may include a Switch Transformer, Conditional Computation in Neural Networks, a Sparse Mixture of Experts, and / or a Megatron-LM.

[0209] Furthermore, an external model (EM) according to one embodiment may refer to a predetermined artificial intelligence model realized by the various disclosed algorithms.

[0210] For example, the external model (EM) may include ChatGPT, Gemini, and / or Llama, etc.

[0211] In embodiments, such external models (EMs) can be selectively used as needed to support the processing of a given task.

[0212] In addition, a specialized model (SM) according to one embodiment may refer to an artificial intelligence model that has undergone optimized learning for a specific purpose, and that has been learned using training data and methods specialized for that purpose.

[0213] That is, a specialized model (SM) may be an artificial intelligence model trained with specialized training data and methods to achieve a predetermined purpose.

[0214] In some embodiments, the specialized model (SM) may include a trained predetermined sLLM (including a MoELM and / or a DMoE model), a general MoE model (NM), and / or an external model (EM), etc. Additionally, the specialized model (SM) may include a specialized module model according to an embodiment disclosed in the "MoE-based Model Identification Method" section below, which provides further details regarding this.

[0215] In an embodiment, the sLLM, general MoE model (NM), external model (EM) and / or specialized model (SM) as described above may be a secondary model (S) that can perform specific tasks through the control and management of the master model (P) of the AI agent model (i.e., orchestrator (OCT) and / or router (RT), etc.).

[0216] -Goal-oriented dialogue method (S100)

[0217] Hereinafter, a detailed description will be given of a goal-oriented dialogue method (S100) that can provide a more accurate response to a user dialogue input by extracting and learning the features of various types of dialogue data sets, generating a dialogue graph based on the extracted features that models predetermined conditional relationships for the dialogue data sets, and selecting the most appropriate dialogue act from among multiple dialogue acts sampled by the already-learned dialogue model in response to the user dialogue input based on the dialogue graph, and providing the selected dialogue act as a response.

[0218] The dialogue dataset may include data relating to dialogues between a variety of speakers occurring in a variety of environments, and therefore includes a variety of types of data depending on the nature of the dialogues occurring between the speakers.

[0219] For example, a first interaction dataset and a second interaction dataset related to different types of tasks may contain different types of data, and the data structure of a first interaction graph generated based on the first interaction dataset and the data structure of a second interaction graph generated based on the second interaction dataset may be different from each other.

[0220] A dialogue graph is a graph that models the relationships between various types of utterances contained in a dialogue dataset and represents them in the form of a graph. It can be structured data on multiple dialogue acts corresponding to the functions, intentions, etc. of multiple utterances, and data on the conditional relationships between multiple dialogue acts.

[0221] The goal-oriented dialogue method (S100) can select and provide an optimal dialogue act group as a response to a user dialogue input from among a plurality of dialogue act groups sampled by a dialogue model based on a dialogue graph.

[0222] In addition, a task requested by the user is determined based on the user interaction input and the interaction act group selected as a response, and the computing system 1000 according to one embodiment can perform the determined task to provide a task execution service to the user.

[0223] The following describes in detail a goal-oriented interaction method (S100) in which a computing system 1000 according to one embodiment provides an appropriate response to a user interaction input based on an interaction graph that models at least one conditional relationship to an interaction dataset, and enables the interaction model to perform a task requested by the user.

[0224] FIG. 9 is a flowchart of a goal-oriented interaction method (S100) according to one embodiment.

[0225] As shown in FIG. 9, a goal-oriented dialogue method (S100) according to one embodiment may include the steps of: generating a dialogue graph (S101) that models at least one conditional relationship for a dialogue dataset; receiving a user dialogue input (S103); sampling a plurality of dialogue act groups for responding to the user dialogue input using a previously trained dialogue model (S105); adjusting the plurality of dialogue act groups based on the dialogue graph (S107); and selecting one of the plurality of dialogue act groups that satisfies a predetermined condition (S109).

[0226] In step (S101), the processors 111, 131 of the system 1000 generate an interaction graph based on various types of interaction datasets.

[0227] For example, the processor 111, 131 generates a first interaction graph that models at least one conditional relationship for a first type of interaction data set associated with a first task, and the processor 111, 131 generates a second interaction graph that models at least one conditional relationship for a second type of interaction data set associated with the first task and another second task.

[0228] Here, the at least one conditional relationship for the dialogue dataset may include at least one of a first conditional relationship (Should relationship) regarding what utterance should be made for an utterance included in the dialogue dataset, a second conditional relationship (Can) regarding what utterance can be made for an utterance, and a third conditional relationship regarding what utterance must not be made for an utterance.

[0229] In step (S103), the processor 111, 131 of the system 1000 receives user interactive input.

[0230] For example, the processors 111, 131 may be communicated with data of user interaction input received via the user input component 121 by the user computing device 110, which may be implemented in a variety of electronic devices.

[0231] In step (S105), the processors 111, 131 of the system 1000 utilize the previously learned dialogue model to sample a plurality of dialogue act groups to provide as responses to the user dialogue input.

[0232] 10, the processors 111, 131 can sample multiple dialogue act groups a1, a2, ... using a dialogue model (π) that has already been trained by the model trainer 160 of the training computing system 150. In this case, the number of multiple dialogue act groups a1, a2, ... sampled by the dialogue model (π) may be several tens, but is not limited to this.

[0233] For example, the first sampled dialogue act group a1 may include four dialogue acts A, B, C, and F, and the second sampled dialogue act group a2 may include three dialogue acts A, C, and G.

[0234] In step (S107), the processors 111, 131 of the system 1000 adjust the sampled dialogue act groups based on the dialogue graph.

[0235] The processor 111, 131 can adjust the plurality of dialogue act groups based on whether each of the plurality of dialogue act groups satisfies at least one conditional relationship included in the dialogue graph in response to a user dialogue input.

[0236] First, the processor 111, 131 determines one of the interaction graphs generated for the various interaction data sets that corresponds to the type of user interaction input.

[0237] The processors 111, 131 then coordinate the plurality of dialogue act groups based on the determined dialogue graph.

[0238] As shown in Figure 10, the determined dialogue graph includes G as a dialogue act that satisfies the first conditional relationship (Should relationship) for the user dialogue input, includes A, C, F, and G as dialogue acts that satisfy the second conditional relationship (Can relationship), and includes A as a dialogue act that satisfies the third conditional relationship (Should-not relationship).

[0239] The processors 111 and 131 remove B from the first dialogue act group a1 so that the first dialogue act group a1 satisfies the second condition relation (Can relation) based on the determined dialogue graph.

[0240] Furthermore, the processors 111 and 131 add G to the first dialogue act group a1 so that the first dialogue act group a1 satisfies the first condition relation (Should relation) based on the determined dialogue graph.

[0241] Furthermore, the processors 111 and 131 remove A from the first dialogue act group a1 so that the first dialogue act group a1 satisfies the third conditional relation (Should-not relation) based on the determined dialogue graph.

[0242] Similarly, the processors 111 and 131 remove A from the second dialogue act group a2 so that the second dialogue act group a2 satisfies the first to third condition relations based on the determined dialogue graph.

[0243] In this case, after the adjustment work for multiple dialogue act groups is completed, the first dialogue act group a1 contains three dialogue acts C, F, and G, and the second dialogue act group a2 contains two dialogue acts C and G.

[0244] By adjusting the sampled dialogue act groups to match the dialogue graph in this way, the reliability of the goal-oriented dialogue service provided by the system 1000 and the controllability over the dialogue model can be improved.

[0245] In step (S109), the processors 111, 131 of the system 1000 select one dialogue act group that satisfies a predetermined condition from the adjusted plurality of dialogue act groups, and provide it as a response to the user dialogue input.

[0246] The processor 111, 131 selects, from the adjusted plurality of dialogue act groups, the dialogue act group that contains the largest number of dialogue acts that satisfy at least one conditional relation in the dialogue graph.

[0247] For example, through an adjustment process, the processors 111 and 131 can select the first dialogue act group a1 that contains the most dialogue acts that satisfy at least one conditional relationship in the dialogue graph from a first dialogue act group a1 containing three dialogue acts C, F, and G and a second dialogue act group a2 containing two dialogue acts C and G, and provide the selected group as a response to the user dialogue input.

[0248] FIG. 11 shows response prediction performance indices (F-1 scores) for user dialogue inputs of various dialogue models (FLAN-T5, GPT-turbo) to which the goal-oriented dialogue method (S100) based on dialogue graphs according to one embodiment of the present disclosure is applied to various dialogue datasets (SGD, MultiWOZ).

[0249] In this case, the response prediction performance index of the dialogue model changes by changing the method of selecting one of multiple dialogue act groups generated from various dialogue models (FLAN-T5, GPT-turbo) and adjusted based on the dialogue graph as a response.

[0250] Furthermore, when selecting one of a plurality of dialogue act groups, the response prediction performance index of the dialogue model also changes depending on whether or not adjustment work is performed based on at least one conditional relation in the dialogue graph.

[0251] For example, the processors 111 and 131 select, from among a plurality of dialogue act groups, a dialogue act group determined by a dialogue model to have the highest response probability, and provide it as a response to a user dialogue input (Greedy).

[0252] The processor 111, 131 also selects, from among the multiple dialogue act groups, the dialogue act group that contains the most dialogue acts that satisfy at least one conditional relationship in the dialogue graph (Compliance), or selects the dialogue act group that is least adjusted based on at least one conditional relationship and provides it as a response to the user dialogue input (Violation).

[0253] Furthermore, the processors 111 and 131 select the dialogue act group that is sampled most frequently from the plurality of dialogue act groups sampled by the dialogue model, and provide it as a response to the user dialogue input (Majority).

[0254] As shown in Figure 11, when multiple dialogue act groups are adjusted based on all of the first conditional relations (Should relations), second conditional relations (Can relations), and third conditional relations (Should-not relations) in the dialogue graph, and the method (Compliance) is followed to select the dialogue act group that contains the most dialogue acts that satisfy at least one conditional relation in the dialogue graph from the multiple dialogue act groups, the response prediction performance index of the dialogue model is the highest.

[0255] In this way, by appropriately adjusting multiple dialogue act groups based on the dialogue graph, and selecting one of the adjusted dialogue act groups that best satisfies the conditional relationships in the dialogue graph and providing it as a response to the user dialogue input, it is possible to provide a more reliable response.

[0256] Meanwhile, the method (S100) may further include a response dialogue act determination step of providing any one of the dialogue acts included in any one of the selected dialogue act groups as a response to the user dialogue input.

[0257] For example, if the user dialogue input is text data in which the speech "Please make a hotel reservation" has been converted into text, the first dialogue act group a1 selected from the adjusted multiple dialogue act groups includes three dialogue acts: "How many people will be staying? (C)", "What date do you wish to make the reservation? (F)", and "What grade of hotel would you like? (G)".

[0258] In the response dialogue act determination step, the processors 111 and 131 select one of the dialogue acts (C, F, G) included in the first dialogue act group a1 to provide it as a response to the user dialogue input. In this case, the processors 111 and 131 can calculate the relevance of each of the dialogue acts (C, F, G) included in the first dialogue act group a1 to the user dialogue input, and can select and provide one of the dialogue acts with the highest relevance as a response to the user dialogue input.

[0259] Furthermore, the processor 111, 131 can determine the type of task requested by the user based on the user interaction input and the interaction act group ultimately selected as a response, and perform the determined task.

[0260] For example, if the task type is determined to be "hotel booking," the processors 111, 131 can complete a hotel booking that meets the user's requirements based on various information related to hotel booking determined in the course of a series of interactions, including user interaction inputs and selected interaction act groups.

[0261] In this way, the system 1000 can provide a more reliable response to the user by adjusting the dialogue acts sampled by the dialogue model based on a dialogue graph in which predetermined condition relationships are structured and modeled based on the dialogue dataset, and providing the adjusted dialogue acts as a response to the user dialogue input.

[0262] Furthermore, the type of task requested by the user is accurately determined based on a series of interactions including a dialogue act adjusted by the user dialogue input and the dialogue graph and determined as the final response according to predetermined conditions, and the system 1000 performs the task thus determined, thereby providing a task execution service that satisfies the user.

[0263] FIG. 12 is a flowchart of a task execution method (S200) based on the context of goal-oriented dialogue through a dialogue model according to one embodiment.

[0264] As shown in FIG. 12, a task execution method (S200) according to one embodiment may include a step of receiving a user dialogue input (S201), a step of determining and providing a response dialogue act for the user dialogue input based on a dialogue graph (S203), a step of analyzing a series of goal-oriented dialogue data including the user dialogue input and the response dialogue act to determine a context of the goal-oriented dialogue (S205), a step of determining a type of task requested by the user based on the context of the goal-oriented dialogue (S207), and a step of performing the task whose type has been determined (S209).

[0265] In step (S201), the processor 111, 131 of the system 1000 receives a user interactive input.

[0266] For example, the processors 111, 131 may be communicated with data of user interaction input received via the user input component 121 by the user computing device 110, which may be implemented in a variety of electronic devices.

[0267] In step (S203), the processor 111, 131 of the system 1000 determines and provides a responsive dialogue act to the user dialogue input based on the dialogue graph associated with the user dialogue input.

[0268] Step (S203) is substantially the same as the goal-oriented dialogue method (S100) described with reference to FIGS. 9 and 10, and therefore a description thereof will be omitted.

[0269] In step (S205), the processors 111, 131 of the system 1000 analyze a series of goal-directed dialogue data including user dialogue inputs and response dialogue acts to determine the context of the goal-directed dialogue.

[0270] The processors 111, 131 of the system 1000 are able to grasp the context of a series of goal-directed dialogues consisting of user dialogue inputs and corresponding responsive dialogue acts.

[0271] For example, the processors 111 and 131 of the system 1000 extract multiple keywords contained in the data of the goal-oriented dialogue, and based on the extracted multiple keywords, analyze the correlations between multiple dialogue acts contained in the dialogue, the intention and purpose of the dialogue, etc., and ultimately determine the context of the goal-oriented dialogue.

[0272] In step (S207), the processor 111, 131 of the system 1000 determines the type of task requested by the user based on the context of the goal-oriented dialogue.

[0273] The processors 111, 131 of the system 1000 determine the type of task corresponding to the determined goal-directed dialogue context based on data related to the task corresponding to the goal-directed dialogue context.

[0274] In this case, task-related data corresponding to the context of the goal-directed dialogue may already be stored in the memory 112, 132 of the system 1000 or may already have been learned by the machine learning model 120, 140.

[0275] For example, the processors 111, 131 of the system 1000 may determine that the task requested by the user is "book a hotel" if it is determined that the context of the goal-oriented dialogue requires an automated hotel reservation service in response to the user's hotel reservation request.

[0276] Meanwhile, the processors 111 and 131 of the system 1000 may determine multiple task types based on the context of the goal-oriented dialogue. The goal-oriented dialogue includes various types of user dialogue inputs and various response dialogue actions thereto, and the context of such goal-oriented dialogue is associated with various tasks. Thus, multiple task types may be determined based on the context of the goal-oriented dialogue according to an embodiment.

[0277] In step (S209), the processors 111 and 131 of the system 1000 perform the task whose type has been determined.

[0278] The processors 111, 131 of the system 1000 execute tasks whose type is determined based on goal-directed dialogue including user dialogue inputs and response dialogue acts, and provide the results of the execution to the user.

[0279] In this case, the processors 111, 131 of the system 1000 generate and execute the programming code necessary to perform the task, the type of which is determined based on the context of the goal-oriented dialogue.

[0280] In addition, in one embodiment, the processors 111, 131 of the system 1000 can receive user interactive input to capture a screen of an electronic device used by the user to obtain a user screen screenshot.

[0281] The processors 111, 131 of the system 1000 then determine the type of task based on information about the determined goal-oriented interaction context and analysis of the user screen shots.

[0282] In this case, the processors 111, 131 of the system 1000 can automatically perform a series of actions (e.g., cursor movement, clicking, text input, etc.) required to perform the task determined on the user screen.

[0283] Furthermore, in one embodiment, when the type of task is determined based on the context of the goal-oriented dialogue, the processors 111, 131 of the system 1000 determine at least one task execution model from among multiple task execution models that is optimized for the task whose type is determined based on the context of the goal-oriented dialogue, and perform the task using the determined at least one task execution model.

[0284] In this case, the method by which the processors 111, 131 of the system 1000 determine at least one task execution model optimized for the task and use it to perform the task is substantially the same as the "MoE-based model identification method" described below, and a description thereof will be omitted here.

[0285] If multiple task types are determined, the processors 111 and 131 of the system 1000 can determine multiple task execution models optimized for each of the multiple tasks and use these to perform the multiple tasks.

[0286] -Method for automatically executing a task determined in response to a user's interactive input (S300)

[0287] FIG. 13 is a flowchart of a method (S300) for automatically executing a task determined in response to user interactive input according to one embodiment.

[0288] According to one embodiment of the method (S300), the processors 111, 131 of the system 1000 can automatically perform tasks requested by the user based on analysis of the user interaction input.

[0289] In this case, when the processors 111 and 131 of the system 1000 perform a task using the method (S300), they obtain the predetermined data required to perform the task based on the user's interactive input, and based on this, the task requested by the user can be automatically performed systematically.

[0290] Furthermore, according to one embodiment of the method (S300), various tasks can be systematically and automatically performed by at least one task execution system determined by simple user interactive input, where the at least one task execution system may include various legacy systems within the enterprise.

[0291] For example, legacy systems may include systems that can support various business operations of an enterprise, such as an Enterprise Resource Planning (ERP) system, a Human Resource (HR) management system, a Customer Relationship Management (CRM) system, a Manufacturing Execution System (MES), and a Quality Management System (QMS).

[0292] As shown in FIG. 13 , a method (S300) according to one embodiment may include a step of receiving a user interaction input (S301), a step of determining a type of task requested by a user based on an analysis result of the user interaction input (S303), a step of obtaining task context data required to perform the task based on the user interaction input (S305), and a step of performing the task based on the task context data (S307).

[0293] In step (S301), the processor 111, 131 of the system 1000 receives a user interactive input.

[0294] For example, the processors 111, 131 may be communicated with data of user interaction input received via the user input component 121 by the user computing device 110, which may be implemented in a variety of electronic devices.

[0295] In step (S303), the processors 111, 131 of the system 1000 perform an analysis of the user interaction input and determine the type of task requested by the user based on the analysis results.

[0296] For example, if the user interaction input is "please make a hotel reservation", the processors 111, 131 of the system 1000 will determine the type of task requested by the user corresponding to the user interaction input as "provide hotel reservation service" through keyword analysis.

[0297] Also, for example, if the user interactive input is "Please approve the expense request for purchasing goods," the processors 111, 131 of the system 1000 can determine that the type of task requested by the user corresponding to the user interactive input is "Approve the expense request."

[0298] Thus, in step (S303), the processors 111, 131 of the system 1000 determine the various task types corresponding to the user interaction input.

[0299] In step (S305), the processor 111, 131 of the system 1000 obtains task context data required to perform the task based on the user interactive input.

[0300] Here, the task context data is data on general basic information for the task execution requested by the user, and may include data on main information required to identify the content of the task.

[0301] The task context data may be included in the data of the user interaction input, or may already be stored in a task context database (not shown) in the form of data of information related to keywords included in the user interaction input.

[0302] For example, if the user interactive input is "Please make a reservation at Hotel A," the task type is determined to be "Providing hotel reservation services," and the information data related to "Hotel A" included in the user interactive input is obtained as task context data as basic information for the execution of the task.

[0303] Also, for example, if the user interactive input is "Please approve the expense request for the purchase of item B," the task type is determined to be "Approval for expense request," and the information data regarding the "expense request for the purchase of item B" included in the user interactive input is obtained as task context data as basic information regarding the execution of the task.

[0304] In this case, data already stored in the task context database, such as information on "specification information for item B," "budget information for the purchase of item B," "past cost information for the purchase of item B," and "approval line policy information," can be obtained as task context data.

[0305] In step (S307), the processors 111 and 131 of the system 1000 execute the task whose type is determined based on the task context data.

[0306] In this case, the processors 111 and 131 of the system 1000 utilize the task context data acquired in step (S305) to perform the task whose type was determined in step (S303).

[0307] For example, if the type of task determined is "approval of expense request," the processors 111 and 131 of the system 1000 identify the task based on information data of "expense request for purchase of item B," and generate data of an electronic approval request for the expense request to purchase item B based on information data such as "specification information of item B," "budget information regarding the purchase of item B," "past expense information regarding the purchase of item B," and "approval line policy information," and provide the data to a user with decision-making authority.

[0308] In this case, the data of the electronic approval request may include a user interface requesting approval of the expense approval for the purchase of item B and general data related to item B. A user with approval authority executes the approval through the user interface requesting approval, and data related to the result of the approval execution is finally transmitted to the user computing device 110 of the user who requested the approval task.

[0309] Meanwhile, in step (S307), the processors 111 and 131 of the system 1000 can perform the task more efficiently by utilizing a task execution system suited to the execution of a particular task.

[0310] For example, as shown in FIG. 14 , step (S307) may include a step (S3071) of determining a task execution system required for task execution based on task type information, and a step (S3073) of executing the task execution system based on task context data to perform the task.

[0311] In step (S3071), the processors 111 and 131 of the system 1000 determine a task execution system suitable for the determined task type.

[0312] For example, the processors 111, 131 of the system 1000 can determine as the task execution system at least one system suitable for the determined type of task from various types of linked legacy systems such as an ERP system, an HR management system, a CRM, an MES, and a QMS.

[0313] In step (S3073), the processors 111, 131 of the system 1000 can control the task to be performed by executing the task execution system determined in step (S3071) based on the task contact data obtained in step (S305).

[0314] In step S307, the processors 111 and 131 of the system 1000 execute the task execution system to perform the task, connect to a predetermined external system, and complete the task execution by performing a predetermined calculation on the external system.

[0315] For example, the processors 111 and 131 of the system 1000 can connect to a predetermined external system by automatically logging in using login information already stored in an authentication information database (not shown). The processors 111 and 131 of the system 1000 can complete a task execution determined by a predetermined operation for the connected external system.

[0316] For example, if the determined task is "order item A," the processors 111 and 131 of the system 1000 connect to the website of the supplier that delivers item A and perform the task of ordering item A. In this case, the processors 111 and 131 of the system 1000 can connect to the supplier's system using login information for the supplier's website and automatically perform the task of ordering item A.

[0317] Additionally, the method (S300) may further include receiving task-related information data related to the task.

[0318] Here, the task-related information data may be data of a user interactive input or data input by a user separately from task context data acquired based on the user interactive input.

[0319] For example, a user may input information about an invoice related to expense processing via the user computing device 110 and request to perform tasks of generating a voucher and requesting approval for the invoice. In this case, data about the invoice related to expense processing may correspond to data about task-related information.

[0320] The task-related information data may also include text data, such as documents, emails, and the like.

[0321] In this case, for example, a user may provide a user dialogue input requesting a task such as "Decide on a meeting date for Project A" together with a report file related to Project A, a document file containing schedule information of collaborators related to Project A, etc. to the system 1000. Here, data for the report file related to Project A, the document file containing schedule information of collaborators related to Project A, etc. may be task-related information data including text data.

[0322] Additionally, the processors 111, 131 of the system 1000 can capture task-related information in text format to obtain task-related information capture data.

[0323] For example, the processors 111 and 131 of the system 1000 may capture at least a portion of a document file provided by a user to obtain task-related information capture data including information about the document file, and may analyze the content of the task-related information using an optical character recognition (OCR) method.

[0324] Additionally, the processors 111, 131 of the system 1000 can generate programming code for performing the task based on analysis of the task-related information capture data, and execute the programming code to perform the task.

[0325] For example, the processors 111, 131 of the system 1000 can perform a task by generating a project A meeting schedule on a date when all collaborators in project A are available, based on an analysis of a document file provided by a user that contains schedule information for collaborators related to project A.

[0326] In this case, the processors 111, 131 of the system 1000 provide the user with task-related recommendation information data based on the results of analyzing various textual task-related information provided by the user.

[0327] Here, the recommended information may include all information that the processors 111, 131 of the system 1000 determine to be useful to the user based on the results of analyzing the task-related information in text format. For example, the recommended information may include information related to document files, related emails, etc., that are related to the task.

[0328] For example, the processors 111 and 131 of the system 1000 can provide the user with recommended information such as the content of an email to be sent to collaborators on Project A and information on materials related to Project A, based on the results of analyzing a report file related to Project A provided by the user, a document file containing schedule information of collaborators related to Project A, etc.

[0329] Thus, according to the method (S300), the processors 111, 131 of the system 1000 can utilize task-related information data provided via the user computing device 110 that is necessary for performing the task in addition to the user interaction input data and task context data to perform the task.

[0330] As a result, the method S300 according to one embodiment allows the processors 111, 131 of the system 1000 to perform tasks based on task context data and / or task-related information data.

[0331] Furthermore, according to other embodiments, the method S300 may further include determining and providing a response dialogue act to the user input based on the dialogue graph, and analyzing data of a series of goal-directed dialogues including the user input and the response dialogue act to determine a context of the goal-directed dialogue.

[0332] Here, the steps of determining and providing a response dialogue act and determining the context of the goal-oriented dialogue are substantially the same as steps S203 and S205 of FIG. 12, and therefore, a description thereof will be omitted here.

[0333] In this case, according to the method S300, the processor 111, 131 of the system 1000, in determining the type of task, can determine the type of task requested by the user based on the context of the goal-oriented dialogue.

[0334] -MoE-based model identification method

[0335] Hereinafter, a method for implementing a model provision service based on a Mixture of Experts (MoE) architecture that realizes modularization for a specific expert model (SM) within an MoE (Mixture of Experts) model in a computing system 1000 according to one embodiment will be described in detail with reference to the accompanying drawings.

[0336] FIG. 15 is a flowchart illustrating an MoE-based model specification method according to an embodiment, and FIG. 16 is a conceptual diagram illustrating an MoE-based model specification method according to an embodiment.

[0337] As shown in Figures 15 and 16, a method for realizing an MoE architecture foundation model providing service in which a computing system 1000 according to one embodiment modularizes a specialized model (SM) including an MoE model may include a step (S401) of performing MoELM-based MoE learning, a step (S403) of acquiring specialized model (SM) characteristic information through MoE learning, a step (S405) of generating a specialized module model based on the acquired specialized model (SM) characteristic information, a step (S407) of acquiring specified domain information, a step (S409) of determining a domain-specific specialized model based on the acquired domain information, a step (S411) of constructing an MoE model based on the determined domain-specific specialized model, and a step (S413) of providing output data based on the constructed MoE model.

[0338] Specifically, in many cases, it is difficult to distinguish or understand the domain that a general, pre-trained specialized model (SM) is specialized for.

[0339] This may result in certain constraints on the selection and utilization of specialized models (SM) optimized for specific tasks.

[0340] To address this issue, in one embodiment, the computing system 1000 can execute the following process to identify the role and / or function of each specialized model (SM) and modularize it, and based on this, effectively select and utilize a customized specialized model (SM) optimized for a specific domain.

[0341] In detail, the computing system 1000 according to one embodiment performs MoELM-based MoE learning (S401).

[0342] That is, in an embodiment, the computing system 1000 can perform MoE learning based on the above-mentioned multiple specialized models (SMs) and router (RT) combination-based MoELM.

[0343] At this time, the learning is performed, and the computing system 1000 can realize learning for each of the multiple specialized models (SM) included in the MoELM.

[0344] In other words, by performing the above-described learning, multiple specialized models (SMs) within the MoELM can be trained individually.

[0345] In other words, the specialized model (SM) according to the embodiment is an artificial intelligence model that has undergone optimized learning for a specific purpose, and may refer to an artificial intelligence model that has been learned using training data and methods specialized for that purpose.

[0346] In embodiments, such specialized models (SM) may include trained predetermined sLLMs (including MoELMs and / or DMoE models), general MoE models (NMs), external models (EMs) and / or specialized module models (MMs) according to embodiments disclosed below.

[0347] In addition, the computing system 1000 according to one embodiment acquires specialized model feature information (SMFI) through MoE learning (S403).

[0348] Here, the specialized model characteristic information (SMFI) according to the embodiment may refer to information that identifies the role and / or function of a specific specialized model (SM).

[0349] In particular, and with further reference to FIG. 8, in an embodiment, the computing system 1000 may further include a Model Specialization Module (MSM) according to one embodiment.

[0350] The computing system 1000 can then obtain the specialized model feature information (SMFI) by working with the model specification module (MSM).

[0351] Here, according to one embodiment, a Model Specialization Module (MSM) may be an artificial intelligence module that generates and outputs Specialized Model Feature Information (SMFI) corresponding to a predetermined Specialized Model (SM) based on MoE learning.

[0352] Specifically, in an embodiment, the model specification module (MSM) can monitor and track the task allocation status of the router (RT) for each specialized model (SM) when the aforementioned MoE learning is performed.

[0353] That is, in an embodiment, the model specification module (MSM) can understand what tasks the router (RT) should allocate and assign to what specialized model (SM) by learning and operating the MoELM.

[0354] According to an embodiment, the model specification module (MSM) may generate a tag for identifying each tracked task assignment state and manage matching.

[0355] This allows the model specification module (MSM) in embodiments to determine expertise for each of a plurality of specialized models (SM).

[0356] In addition, in the embodiment, the model specification module (MSM) generates specialized model feature information (SMFI) corresponding to each specialized model (SM) based on the determined specialized model (SM)-specific specialized model (SM).

[0357] FIG. 17 is a diagram illustrating an example of specialized model feature information (SMFI) according to one embodiment.

[0358] Here, as shown in FIG. 17, in an embodiment, the model specification module (MSM) can generate the specialized model characteristic information (SMFI) in at least one of the following forms:

[0359] [Form 1] Specialist model characteristic information (SMFI) in the form of selecting one of the specialist model (SM) role and / or function specific categories (e.g., Q&A or device control) already set according to user input.

[0360] [Second Form] Specialist Model Feature Information (SMFI) in a natural language form that identifies the role and / or function of the Specialist Model (SM)

[0361] [Form 3] Specialized Model Characterization Information (SMFI) in a form that identifies the role and / or function of the Specialized Model (SM) in at least one of the Forms 1 and 2 and further defines the input data and output data of the Specialized Model (SM).

[0362] Subsequently, in an embodiment, the model specification module (MSM) may provide the generated specialized model characteristic information (SMFI) to the computing system 1000 as output data.

[0363] Therefore, in the embodiment, the computing system 1000 can acquire characteristic information for each specialized model (SM) by interfacing with the model specification module (MSM).

[0364] Furthermore, the computing system 1000 according to one embodiment generates a specialized module model (MM) based on the acquired specialized model characteristic information (SMFI) (S405).

[0365] Here, the specialized module model (MM) according to one embodiment may refer to a specialized model (SM) that is matched with predetermined specialized model characteristic information (SMFI) and is independently separated.

[0366] In particular, in an embodiment, the computing system 1000 matches the specialized model characteristic information (SMFI) obtained as described above to a corresponding specialized model (SM).

[0367] In addition, in the embodiment, the computing system 1000 independently separates and stores the specialized models (SM) matched with the specialized model characteristic information (SMFI) into a database.

[0368] That is, in the embodiment, the computing system 1000 performs modularization by matching each specialized model (SM) with corresponding specialized model characteristic information (SMFI), and separately storing and managing them.

[0369] Therefore, the computing system 1000 can generate a specialized module model (MM) that is an independently separated specialized model (SM) while matching the specialized model characteristic information (SMFI).

[0370] Thus, in an embodiment, the computing system 1000 understands the characteristics of each specialized model (SM) within a given MoE model (in an embodiment, MoELM) and modularizes each specialized model (SM) to a small size that allows it to be reused and shared.

[0371] This enables the computing system 1000 to quickly and efficiently select and sort specialized models (SMs) that realize data processing processes optimized for specific domains with higher accuracy, and easily support flexible expansion or contraction of MoE models based on this.

[0372] Additionally, the computing system 1000 according to one embodiment may acquire predetermined domain information (S407).

[0373] Here, domain information according to an embodiment may be information that defines a domain that identifies the data, rules, terminology, problem definitions, and / or processes that a given AI system uses to perform a given task.

[0374] In particular, in an embodiment, the computing system 1000 acquires predetermined input data (eg, text, audio, image, video, and / or specific sensor-based sensing data, etc.).

[0375] Additionally, in an embodiment, the computing system 1000 determines a domain corresponding to the obtained input data.

[0376] Here, in an embodiment, the method by which the computing system 1000 determines the domain for input data may be performed based on various disclosed algorithms that can perform this, and the embodiment of the present invention does not limit or restrict the algorithm itself.

[0377] Therefore, the implementation system 1000 can obtain domain information for the task to be processed.

[0378] Additionally, the computing system 1000 according to one embodiment determines a domain-specific expert model of the acquired domain information base (S409).

[0379] Here, a domain-specific specialist model according to an embodiment may refer to a specialist model (SM) that performs data processing (e.g., deep learning) operations optimized for a specific domain.

[0380] In particular, and with further reference to FIG. 16, in an embodiment, the computing system 1000 determines at least one domain-specific specialty model based on the domain information and specialty model characteristic information (SMFI) obtained as described above.

[0381] More specifically, in an embodiment, the computing system 1000 may detect at least one specialized model characteristic information (SMFI) having characteristics corresponding to the obtained domain information.

[0382] For example, when the computing system 1000 confirms the "characteristics of the task of outputting response data to specified question and answer data" based on the first domain information, it can detect at least one specialized model characteristic information (SMFI) that is identified as a "role and / or function specialized for question and answering" from among multiple specialized model characteristic information (SMFI) that has been databased.

[0383] Here, according to an embodiment, the computing system 1000 can detect at least one specialized model characteristic information (SMFI) corresponding to domain information based on multiple tags generated by the model specification module (MSM) for each task allocation state of the router (RT) for multiple specialized models (SM) during the aforementioned MoE architecture-based learning.

[0384] That is, according to an embodiment, the computing system 1000 can detect at least one specialized model characteristic information (SMFI) corresponding to the corresponding domain information by comparing the multiple tags and domain information generated as described above.

[0385] Here, according to an embodiment, the computing system 1000 filters tags to be compared according to the time at which each tag was generated.

[0386] Specifically, the computing system 1000 sets at least one tag generated at a specific task assignment time as a comparison target tag according to user input and / or a pre-established unique process.

[0387] For example, the computing system 1000 may set at least one tag generated for a task assignment state that has occurred since a previously set time during the entire learning time as a comparison tag, focusing on the fact that the higher the learning rate, the more accurate the task assignment.

[0388] Therefore, the computing system 1000 can detect at least one specialized model characteristic information (SMFI) corresponding to the domain information by comparing at least one filtered tag and the domain information, which ensures higher accuracy.

[0389] In addition, in an embodiment, the computing system 1000 extracts a specialized model (SM) (ie, specialized module model (MM)) that is matched to each of the detected at least one specialized model characteristic information (SMFI).

[0390] Then, in an embodiment, the computing system 1000 determines the extracted at least one specialized module model (MM) as a domain-specific specialized model.

[0391] Furthermore, the computing system 1000 according to an embodiment of the present invention constructs an MoE model based on the determined domain-specific expert model (S411).

[0392] Further referring to FIG. 16, in an embodiment, the computing system 1000 can construct a model (hereinafter referred to as a DMoE model) that operates like an MoE architecture based on at least one domain-specific specialized model determined as described above.

[0393] In other words, the computing system 1000 can construct an MoE model (i.e., a DMoE model) that realizes data processing optimized for a specific domain by utilizing at least some of the small-sized, modularized specialized models (SMs) (i.e., domain-specific specialized models).

[0394] In particular, in an embodiment, the computing system 1000 can combine at least one domain-specific specialized model and a predetermined router (RT) to construct the aforementioned DMoE model.

[0395] Thus, in an embodiment, the computing system 1000 can build a DMoE model that includes domain-specific specialized models and routers (RTs).

[0396] Here, depending on the embodiment, the DMoE model may be included in the sLLM according to the embodiment of the present invention.

[0397] In other words, an sLLM according to an embodiment may include a DMoE model constructed according to one embodiment.

[0398] Furthermore, the computing system 1000 according to an embodiment of the present invention provides output data based on the constructed MoE model (S413).

[0399] That is, in an embodiment, the computing system 1000 uses the DMoE model constructed as described above to provide output data (e.g., response data to a specific question and / or a control signal based on a specific command word) for specified input data (e.g., text, voice, image, video, and / or sensing data based on a specific sensor, etc.).

[0400] As described above, in an embodiment, the computing system 1000 can identify the role and / or function of each specialized model (SM) while simultaneously separating and modularizing them to a level that allows them to be reused and shared. This can be utilized to quickly and flexibly construct a customized MoE model (i.e., a DMoE model) optimized for a specific domain, and provide specified output data through efficient task processing using the constructed model.

[0401] In other words, in an embodiment, the computing system 1000 can realize and provide an MoE model with further improved data processing (and / or calculation) speed and inference performance, thereby supporting a variety of services, thereby effectively improving their performance and quality.

[0402] -MoE-based LLM-based AI agent provision method

[0403] Hereinafter, a method for implementing a model provision service based on MoE architecture in which a computing system 1000 according to an embodiment of the present invention determines an application model optimized for a domain according to an external environment based on a Large Language Model (LLM) applying MoE (Mixture of Experts) and provides an on-device specialized AI agent (Artificial Intelligence Agent) that executes output based on the determined application model will be described in detail with reference to the accompanying drawings.

[0404] Figure 18 is a flowchart for explaining a method for providing an AI agent based on an MoE application LLM according to one embodiment, and Figure 19 is a conceptual diagram for explaining a method for providing an AI agent based on an MoE application LLM according to one embodiment.

[0405] As shown in Figures 18 and 19, a method for implementing an MoE architecture-based model provision service in which a computing system 1000 according to one embodiment determines an application model optimized for a domain based on an external environment based on an LLM applying MoE and provides an on-device specialized AI agent specialized model (AIAM) that performs output based on the determined application model may include a step of executing an on-device AI agent service (S501), a step of acquiring specified input data (S503), a step of determining a domain based on the acquired input data (S505), a step of determining an application model based on the determined domain (S507), and a step of providing output data based on the determined application model (S509).

[0406] Specifically, the computing system 1000 according to an embodiment of the present invention executes an on-device AI agent service (S501).

[0407] For reference, on-device AI may refer to technology that performs artificial intelligence-based data processing directly within a user's device, rather than on the cloud and / or an external server. This can provide advantages such as privacy protection, real-time processing, and reduced dependency on internet connections, as all processing is completed within the device without sending data externally.

[0408] Therefore, in this context, on-device AI agent services may refer to various services realized by utilizing on-device AI.

[0409] Illustratively, on-device AI agent services may include smartphone voice assistant services (e.g., Google Assistant, Apple Siri, Samsung Bixby, etc.), smart camera services (e.g., Google Pixel's HDR+, Apple's Deep Fusion, etc.), fitness tracker and smartwatch services (e.g., Apple Watch, Fitbit, etc.), automobile self-driving services (e.g., Tesla's Autopilot, etc.), and / or home security services (e.g., Nest Secure, Ring, etc.).

[0410] In an embodiment, the computing system 1000 may execute a predetermined on-device AI agent service based on an AI agent specialization model (AIAM) according to one embodiment and / or in conjunction with a predetermined application, etc.

[0411] Furthermore, the computing system 1000 according to an embodiment of the present invention acquires predetermined input data (S503).

[0412] In detail, in an embodiment, the computing system 1000 can obtain at least one input data (e.g., predetermined text, audio, image, video, and / or sensing data, etc.) based on user input and / or interaction with an external device (e.g., predetermined sensor, etc.) based on the on-device AI agent service executed as described above.

[0413] In an embodiment, the input data obtained as described above may include predetermined data that can identify a target task for data processing.

[0414] Furthermore, the computing system 1000 according to one embodiment determines a domain based on the acquired input data (S505).

[0415] In other words, a domain according to an embodiment may refer to the data, rules, terminology, problem definition and / or process, etc., that a given AI system uses to perform a particular task.

[0416] In particular, in an embodiment, the computing system 1000 determines a domain corresponding to the obtained input data.

[0417] Here, in an embodiment, the method by which the computing system 1000 determines the domain for input data may be performed based on various disclosed algorithms that can perform this, and in one embodiment, the algorithm itself is not limited or restricted.

[0418] Therefore, in an embodiment, the computing system 1000 can obtain domain information corresponding to the task to be processed.

[0419] Furthermore, the computing system 1000 according to one embodiment determines an application model according to the determined domain (S507).

[0420] Here, the application model according to the embodiment may refer to a model that performs a predetermined task process using given input data.

[0421] In an embodiment, such an applied model may be at least one of the secondary models (S) as described above.

[0422] In other words, the secondary model (S) according to the embodiment may refer to a model that can perform a specific task through the control and management of a master model (P) (i.e., an orchestrator (OCT) and / or a router (RT), etc.) that is responsible for controlling and managing the operation of a given AI system.

[0423] In embodiments, such secondary models (S) may include at least one of the following models: sLLM (including MoELM and / or DMoE models), general MoE models (NM), external models (EM) and / or specialized models (SM) (including specialized module models (MM)).

[0424] In particular, in an embodiment, the computing system 1000 determines at least one application model based on the domain information obtained as described above.

[0425] More specifically, in one embodiment, the computing system 1000 works in conjunction with a master model (P) (i.e., an orchestrator (OCT) and / or a router (RT), etc.) in one embodiment to detect at least one model (i.e., a domain-specific model) among the secondary models (S) that performs data processing (e.g., deep learning, etc.) operations optimized for given domain information.

[0426] Here, in one embodiment, the specific method by which the computing system 1000 detects a domain-specific model in conjunction with the master model (P) will be omitted, as the explanation regarding the router (RT) and orchestrator (OCT) disclosed in the aforementioned "AI Agent Specialization Model (AIAM)" will apply mutatis mutandis.

[0427] Additionally, in an embodiment, the computing system 1000 determines the detected at least one domain-specific model as an application model.

[0428] Additionally, the computing system 1000 according to one embodiment provides output data based on the determined application model (S509).

[0429] That is, in an embodiment, the computing system 1000 can generate and provide output data (e.g., response data to a specific question and / or a control signal based on a specific command word, etc.) for predetermined input data (e.g., text, voice, image, video, and / or sensing data based on a specific sensor, etc.) based on at least one application model determined through the AI Agent Specialization Model (AIAM) as described above.

[0430] In other words, the computing system 1000 executes a predetermined requested task based on given input data using the application model determined as described above, and provides output data resulting from the executed data processing.

[0431] Here, in an embodiment, the computing system 1000 can provide the output data based on the on-device AI agent service described above.

[0432] As described above, in the embodiment, the computing system 1000 can effectively determine a model optimized for data processing according to a given domain even in an on-device environment based on an AI Agent Specialization Model (AIAM) including a model realized by applying the MoE architecture in various embodiments (e.g., MoELM, DMoE model, and / or Specialized Module Model (MM)), and provide output based on efficient data processing through the determined model.

[0433] That is, the computing system 1000 can implement and provide an artificial intelligence model (i.e., an AI Agent Specialization Model (AIAM)) that better hears, understands, performs, and responds to a given task in any environment.

[0434] Thus, in an embodiment, the computing system 1000 can directly and significantly improve the quality and performance of a variety of AI agent-based services (e.g., smartphone voice assistant services, smart camera services, fitness tracker and smart watch services, self-driving automobile services, and / or home security services).

[0435] The above-described embodiments of the present invention may be embodied in the form of program instructions executable by various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include, alone or in combination, program instructions, data files, data structures, and the like. The program instructions recorded on the computer-readable recording medium may be specially designed or constructed for the present invention, or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include not only machine language code, such as produced by a compiler, but also high-level language code executable by a computer using an interpreter, for example. A hardware device may be replaced by one or more software modules to perform processes according to the present invention, and vice versa.

[0436] The specific implementation described in the present invention is one embodiment and does not limit the scope of the present invention in any way. For the sake of brevity, descriptions of conventional electronic configurations, control systems, software, and other functional aspects of the system may be omitted. Furthermore, line connections or connecting members between components shown in the drawings are illustrative of functional connections and / or physical or circuit connections, and may be represented as various alternative or additional functional connections, physical connections, or circuit connections in an actual device. Furthermore, unless specifically referred to as "essential," "critically," etc., a component may not be essential for application of the present invention.

[0437] Furthermore, although the detailed description of the present invention has been described with reference to preferred embodiments of the present invention, it will be understood by those skilled in the art or those having ordinary knowledge in the art that various modifications and changes can be made to the present invention without departing from the spirit and technical scope of the present invention as set forth in the claims below. Therefore, the technical scope of the present invention should not be limited to the content of the detailed description of the specification, but should be defined by the claims. [Explanation of symbols]

[0438] 10 Interaction graph generation module 20 Dialogue Act Group Sampling Module 30 Dialogue Action Group Coordination Module 40 Dialogue Act Group Selection Module 50 Dialogue Act Selection Module 110, 180, 181, 182, 183 User computing devices 111, 131, 151 processors 112, 132, 152 memory 113, 133, 153 data 114, 134, 154 imperative words 120, 140 Machine Learning Models 121 User Input Components 160 Model Trainer 161 training data 100, 200, 400 computing devices 300 Neuromorphic Circuits 310 Presynaptic Neuron Circuits 311 Presynaptic Line 320 Postsynaptic Neuronal Circuits 330 Synaptic Circuits

Claims

1. 1. A method for automatically executing a task determined in response to user input by a computing system including a memory and a processor, comprising: receiving user interactive input; determining a type of task requested by the user based on the analysis of the user interaction input; obtaining task context data necessary to perform the task based on the user interaction input; and performing said task based on said task context data.

2. In the step of performing the task, determining a task execution system required to execute the task based on the task type information; 2. The method of claim 1, further comprising: executing the task execution system based on the task context data to perform the task.

3. receiving task-related information data relating to the task; In the step of performing the task, 2. The method for automatically executing a task determined in response to a user interactive input according to claim 1, wherein the task is executed based on at least one of the task context data and the task-related information data.

4. the task-related information data includes text data; In the step of performing the task, capturing the task-related information in text format to obtain task-related information capture data; generating programming code for performing the task based on an analysis of the task-related information capture data; 4. The method for automatically performing a task determined in response to user interactive input of claim 3, further comprising executing said programming code to perform said task.

5. When executing the task execution system to perform the task, 3. The method for automatically executing a task determined in response to a user interactive input according to claim 2, further comprising automatically connecting to a predetermined external system and completing the execution of the task by performing a predetermined operation on the predetermined external system.

6. determining and providing a responsive dialogue act to the user input based on the dialogue graph; and analyzing a series of goal-directed dialogue data including the user input and the response dialogue act to determine a context of the goal-directed dialogue; determining a type of task, 2. The method for automatically executing a task determined in response to a user interaction input according to claim 1, further comprising determining the type of task requested by the user based on the context of the goal-oriented interaction.

7. determining a context for the goal-directed dialogue; 7. The method for automatically executing a task determined in response to a user dialogue input according to claim 6, further comprising: extracting a plurality of keywords from the data of the goal-oriented dialogue; and analyzing at least one of the correlations between a plurality of dialogue acts included in the goal-oriented dialogue, the intention of the goal-oriented dialogue, and the purpose of the goal-oriented dialogue based on the plurality of keywords to determine the context.

8. determining at least one task execution model optimized for the task, the type of which is determined based on the context of the goal-oriented dialogue, from among a plurality of task execution models; 7. The method for automatically executing a task determined in response to a user interactive input according to claim 6, wherein the step of performing the task uses the determined at least one task execution model to perform the task.

9. The steps of determining a type of task and performing the task include:

2. The method of claim 1, wherein the computing system determines the type of the task and supports predetermined operations required to perform the task.

10. In the step of performing the task, conducting an analysis of programming code associated with said task; generating programming code for performing the task based on the analysis; 2. The method for automatically performing a task determined in response to user interactive input of claim 1, further comprising executing said programming code to perform said task.

11. receiving a user interactive input to capture a screen of the electronic device utilized by the user to obtain a user screen screenshot; 7. The method for automatically executing a task determined in response to user interaction input according to claim 6, wherein the step of determining the type of task determines the type of task based on information about the determined context of the goal-oriented interaction and an analysis of the user screen screenshots.

12. The step of determining and providing a response dialogue act comprises: generating an interaction graph modeling at least one conditional relationship for the interaction dataset; sampling a plurality of dialogue act groups for responding to the user dialogue input using a previously trained dialogue model; adjusting the plurality of dialogue act groups based on the dialogue graph; and 7. The method for automatically executing a task determined in response to a user dialogue input according to claim 6, further comprising the step of selecting one of the plurality of dialogue act groups that satisfies a predetermined condition.

13. 13. The method for automatically executing a task determined in response to a user dialogue input according to claim 12, wherein the at least one conditional relation includes at least one of a first conditional relation regarding what utterance should be made for one utterance in a dialogue flow, a second conditional relation regarding what utterance can be made for one utterance, and a third conditional relation regarding what utterance cannot be made for one utterance.

14. In the step of selecting any one of the dialogue act groups, The method for automatically executing a task determined in response to a user dialogue input according to claim 12, further comprising the step of selecting one of the dialogue act groups that best satisfies the at least one conditional relation.

15. at least one memory; and at least one processor that reads at least one instruction stored in the memory to perform a task execution method based on the context of a goal-directed dialogue; The at least one processor receiving user interactive input; determining a type of task requested by the user based on the analysis of the user interaction input; obtaining task context data necessary to perform the task from data associated with the user interactive input; A system for automatically executing tasks determined in response to user interactive input, performing the tasks based on the task context data.

16. an electronic device for receiving user interactive input; and a computing device including at least one memory and at least one processor that reads at least one instruction word stored in the at least one memory to perform a task execution method based on a context of a goal-directed interaction; The at least one processor receiving user interactive input; determining a type of task requested by the user based on the analysis of the user interaction input; obtaining task context data necessary to perform the task based on the user interactive input; A system for automatically executing tasks determined in response to user interactive input, performing the tasks based on the task context data.

Images