K-GPT solution-based electronic device including generative artificial intelligence working with external model and method for controlling electronic device

The integration of internal and external AI models in an electronic device addresses the limitations of existing chatbots by providing personalized and secure responses, enhancing the AI's capabilities and service range.

JP2024164819AActive Publication Date: 2024-11-27PERSONA AI CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
JP2024077267
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-06-27
Filing Date
2024-05-10
Publication Date
2024-11-27
Estimated Expiration
2044-05-10

AI Technical Summary

Technical Problem

Existing artificial intelligence models, such as chatbots, are trained on large-scale databases and struggle to provide personalized responses due to a lack of specialization and security vulnerabilities when used by multiple users, while training on individual data limits their universal problem-solving ability.

Method used

An electronic device integrates both an internal and external AI model, where the internal model provides personalized responses based on user-specific data and the external model offers general-purpose solutions, with security measures to prevent information leakage.

Benefits of technology

The system provides tailored responses while ensuring security, enhancing the AI's problem-solving ability and range of services, offering specialized and universal responses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024164819000001_ABST
    Figure 2024164819000001_ABST
Patent Text Reader

Abstract

To provide a K-GPT solution-based electronic device including generative artificial intelligence that works with an external model, and a method for controlling the electronic device.SOLUTION: An electronic device 100 includes: a memory 110 that includes a first artificial intelligence model 111 that performs query-reply; a communication unit 120 that performs communication with an external electronic device including a second artificial intelligence model that performs the query-reply; and a processor 130 that provides a reply to a user-query via at least one of the first artificial intelligence model and the second artificial intelligence model.SELECTED DRAWING: Figure 6
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present disclosure relates to an electronic device, and more particularly, to an electronic device that provides various tasks and functions by utilizing both its own chatbot and external chatbots. [Background technology]

[0002] Recently, various generative AIs, including Chat-GPT, have been developed and are being offered as services. In particular, AI models such as chatbots that provide various information and perform tasks in a dialogue format with users have a major advantage in that they can communicate directly with users.

[0003] However, existing AI models such as chatbots are often trained on large databases and then shared among many users, which leaves them lacking in the ability to specialize for personal information or business purposes.

[0004] Conversely, when training an AI model based solely on information / data held by an individual, there is the problem that it is difficult to significantly improve the problem-solving ability of the AI ​​model in a universal direction, and there is also the disadvantage that the range of information handled is too narrow. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Korean Patent No. 10-2436549 [Patent Document 2] Korean Patent Publication No. 10-2022-0046964 [Patent Document 3] Korean Patent Publication No. 10-2019-0055425 Summary of the Invention [Problem to be solved by the invention]

[0006] The present disclosure provides an electronic device that can utilize an artificial intelligence model stored in an external electronic device while proceeding with learning based on information sensitive to or specialized tasks for the artificial intelligence model stored within itself, thereby utilizing internal and external artificial intelligence models in a mutually complementary manner to provide appropriate responses and business results.

[0007] In particular, the present disclosure provides electronic devices that leverage internal and external artificial intelligence models to provide responses / services that are highly tailored to the individual user or the organization / institution to which the user belongs, while still being non-vulnerable in terms of security.

[0008] The object of the present disclosure is not limited to the above object, and other objects and advantages of the present disclosure not mentioned can be understood from the following description and will be more clearly understood by the embodiments of the present disclosure. Furthermore, it can be easily understood that the objects and advantages of the present disclosure can be realized by the means and combinations thereof shown in the claims. [Means for solving the problem]

[0009] An electronic device according to an embodiment of the present disclosure includes a memory including a first artificial intelligence model for performing question and answering, a communication unit for performing communication with an external electronic device including a second artificial intelligence model for performing question and answering, and a processor for providing a response to a user's question through at least one of the first artificial intelligence model and the second artificial intelligence model. When a user's question for at least one document is acquired, the processor acquires a response of the first artificial intelligence model to the question, updates the question based on the response of the first artificial intelligence model, and acquires a response of the second artificial intelligence model based on the updated question.

[0010] The processor may classify a category of the document based on a semantic analysis of the first artificial intelligence model for the document and store the document in a knowledge base associated with the electronic device based on the classified category.

[0011] In this case, the processor can obtain a response of the first artificial intelligence model to the user's question based on at least one target document in the knowledge base that is in the same category as the document.

[0012] Further, the processor may identify at least one keyword included in the user's question, identify a plurality of documents in the knowledge base that are included in the same category as the document, select at least one target document from the plurality of documents that is related to the keyword, and obtain a response of the first artificial intelligence model to the user's question based on the selected target document.

[0013] Meanwhile, in the process of updating the question, the processor may replace personal information contained in at least one of the user's question and the response of the first artificial intelligence model with example information.

[0014] In addition, when a security level of a certain level or higher is set for at least one of a first document related to the question and a second document related to the response of the first artificial intelligence model, the processor may identify the security information included in the response of the first artificial intelligence model, and in the process of updating the question, replace the security information included in at least one of the user's question and the first artificial intelligence model with exemplary information.

[0015] The artificial intelligence model may include at least one of an engine including a recurrent neural network (RNN) model, a natural language generation engine, a natural language processing engine, a natural language understanding engine, and a dialogue engine, and the processor may train at least one engine constituting the artificial intelligence model based on the document.

[0016] The processor can identify at least one task requested according to the user's question, and if the identified task is included in a pre-defined first task, obtain a response from a first artificial intelligence model to the question, update the question based on the response of the first artificial intelligence model, obtain and provide a response from the second artificial intelligence model based on the updated question, and if the identified task is included in a pre-defined second task, obtain and provide the response from the first artificial intelligence model to the question.

[0017] Here, when a user input requesting an additional response is received after the identified task is included in the second task and the response of the first artificial intelligence model is provided as the first response, the processor may update the question based on the response of the first artificial intelligence model, obtain the response of the second artificial intelligence model based on the updated question, and provide it as the second response. When the second response is provided, the processor may identify whether there is additional information included in the second response but not included in the first response, and if there is additional information, update the first task and the second task so that the identified task is included in the first task that is not the second task, and if there is no additional information, train the first artificial intelligence model based on the second response.

[0018] A control method of an electronic device including a first artificial intelligence model for performing question-and-answer processing according to one embodiment of the present disclosure includes a step in which, when a user question for at least one document is obtained, the electronic device obtains a response of the first artificial intelligence model to the question, the electronic device updates the question based on the response of the first artificial intelligence model, and the electronic device obtains a response of a second artificial intelligence model that is stored in an external electronic device and performs question-and-answer processing based on the updated question. Effect of the Invention

[0019] The electronic device and control method according to the present disclosure can selectively utilize internal and external artificial intelligence models to provide responses that are specialized for a user, such as personal or security documents, while at the same time providing a wide range of universal and useful responses / services. [Brief description of the drawings]

[0020] [Figure 1A] FIG. 1 is a block diagram illustrating a configuration of an electronic device according to an embodiment of the present disclosure. [Figure 1B] FIG. 1 is a block diagram illustrating the operation of an electronic device that uses internal and external artificial intelligence models to provide responses according to one embodiment of the present disclosure. [Diagram 2] 1 is a flowchart for explaining an operation of an electronic device according to an embodiment of the present disclosure. [Diagram 3] FIG. 2 is a block diagram illustrating the functional configuration of an electronic device that uses a self-stored artificial intelligence model and an external artificial intelligence model according to an embodiment of the present disclosure. [Figure 4] 1 is an algorithm for explaining an operation of an electronic device according to an embodiment of the present disclosure selectively utilizing an artificial intelligence model depending on the type of requested task. [Figure 5A] FIG. 2 illustrates an embodiment of an electronic device according to the present disclosure that accepts text input along with a user question and provides one or more responses to the question. [Figure 5B] FIG. 2 illustrates an embodiment of an electronic device according to the present disclosure that accepts text input along with a user question and provides one or more responses to the question. [Figure 6] FIG. 1 is a block diagram illustrating a configuration of an electronic device according to various embodiments of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Before describing the present disclosure in detail, the method of describing the present specification and drawings will be described.

[0022] First, the terms used in this specification and the claims are general terms selected in consideration of the functions in various embodiments of the present disclosure. However, such terms may change depending on the intentions of engineers in the relevant technical field, legal or technical interpretations, and the emergence of new technologies. In addition, some terms are terms arbitrarily selected by the applicant. Such terms may be interpreted as defined in this specification, and unless a specific term is defined, they may be interpreted based on the overall content of this specification and common general technical knowledge in the relevant technical field.

[0023] Furthermore, the same reference numbers or symbols in the drawings attached to this specification indicate parts or components that perform substantially the same functions. For the sake of convenience in explanation and understanding, the same reference numbers or symbols are used in different embodiments. In other words, even if components having the same reference numbers are all shown in multiple drawings, multiple drawings do not mean one embodiment.

[0024] In addition, in the present specification and claims, terms including ordinal numbers such as "first", "second", etc. may be used to distinguish between components. Such ordinal numbers are used to distinguish between identical or similar components, and the meaning of a term should not be interpreted in a limiting manner due to the use of such ordinal numbers. For example, the order of use, arrangement, etc. of components combined with such ordinal numbers should not be limited by the number. If necessary, each ordinal number may be used interchangeably.

[0025] In this specification, the singular includes the plural unless the context clearly indicates otherwise. In this application, the terms "comprise" or "comprise" are intended to specify the presence of features, numbers, steps, operations, components, parts, or combinations thereof described herein, and should be understood as not precluding the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0026] In the embodiments of the present disclosure, terms such as "module," "unit," "part," and the like are terms for indicating components that perform at least one function or operation, and such components may be realized in hardware or software, or a combination of hardware and software. Furthermore, a plurality of "modules," "units," "parts," and the like may be integrated into at least one module or chip and realized as at least one processor, except when each of them needs to be realized in separate specific hardware.

[0027] In the embodiments of the present disclosure, when a part is connected to another part, this includes not only a direct connection but also an indirect connection via another medium. Furthermore, unless otherwise specified, when a part includes a certain component, it does not exclude other components, but means that the part may further include other components.

[0028] 1A is a block diagram illustrating a configuration of an electronic device according to an embodiment of the present disclosure. Referring to FIG. 1A, an electronic device 100 includes a memory 110, a communication unit 120, and a processor .

[0029] The electronic device 100 may be implemented as a server including one or more computers, and may also be implemented as a smartphone, a tablet PC, a desktop PC, or the like.

[0030] The memory 110 is configured to store an operating system (OS) for controlling the overall operation of the components of the electronic device 100 and at least one instruction or data related to the components of the electronic device 100.

[0031] The memory 110 can include a non-volatile memory such as a ROM or a flash memory, and can include a volatile memory such as a DRAM. The memory 110 can also include a hard disk, a solid state drive (SSD), and the like.

[0032] The communication unit 120 is a component that allows the electronic device 100 to communicate with various external electronic devices. For example, the electronic device 100 may communicate with one or more sensor devices, gateways, etc., via the communication unit 120. When the electronic device 100 is a server, the electronic device 100 may communicate with user terminals of various users and provide various UI / UX related to services via the user terminals.

[0033] The communication unit 120 may include a circuit, a module, a chip, etc. for communicating with at least one external device through various wired and wireless communication methods. The communication unit 120 may be connected to an external device through various networks.

[0034] Depending on the area or scale, the network may be a personal area network (PAN), a local area network (LAN), a wide area network (WAN), etc., and depending on the openness of the network, it may be an intranet, an extranet, or the Internet, etc.

[0035] The communication unit 120 may be connected to an external device via various wireless communication methods such as LTE (long-term evolution), LTE-A (LTE Advance), 5G (5th Generation) mobile communication, CDMA (code division multiple access), WCDMA (wideband CDMA), UMTS (universal mobile telecommunications system), WiBro (Wireless Broadband), GSM (Global System for Mobile Communications), DMA (Time Division Multiple Access), WiFi (Wi-Fi), WiFi Direct, Bluetooth (registered trademark), BLE (Bluetooth Low Energy), NFC (near field communication), Zigbee (registered trademark), and LoRa.

[0036] The communication unit 120 can also be connected to an external device via a wired communication method such as Ethernet, an optical network, a Universal Serial Bus (USB), or ThunderBolt.

[0037] In addition, the communication unit 120 may be configured using various communication methods / technologies that will be newly invented in the future.

[0038] The processor 130 is configured to generally control the electronic device 100. Specifically, the processor 130 is connected to the memory 110 and can execute at least one instruction stored in the memory 110 to perform operations according to various embodiments of the present disclosure.

[0039] The processor 130 may include a general-purpose processor such as a CPU, AP, or DSP (Digital Signal Processor), a dedicated graphics processor such as a GPU or VPU (Vision Processing Unit), or a dedicated artificial intelligence processor such as an NPU, etc. The dedicated artificial intelligence processor may be designed with a hardware structure specialized for training or using a specific artificial intelligence model.

[0040] FIG. 1B is a block diagram illustrating the operation of an electronic device that uses internal and external artificial intelligence models to provide responses according to one embodiment of the present disclosure.

[0041] Referring to FIG. 1B, the electronic device 100 can include a first artificial intelligence model 111 for performing question-answering.

[0042] The artificial intelligence model 111 may include various engines for natural language processing and dialogue progression (e.g., DNN / RNN Engine, NLG, NLU, Dialog Engine, morphological analysis engine, individual analysis engine, semantic analysis engine, speech line analysis engine, related / similar word analysis engine, user-defined dictionary, etc.). The artificial intelligence model 111 may include a generative model related to dialogue, etc. (e.g., Generative Pre-trained Transformer). The first artificial intelligence model 111 according to the present disclosure may be referred to as "K-GPT" and may proceed with communication with a user based on the concept of a chatbot that executes dialogue.

[0043] The artificial intelligence model 111 may be stored in the form of data in the memory 110 of the electronic device 100, but may also be realized at least in part in hardware under the control of the processor 130.

[0044] 1B, the electronic device 100 may communicate with at least one external electronic device 200 via the communication unit 120. The external electronic device 200 may correspond to at least one server or terminal device. For example, the external electronic device 200 may include a second artificial intelligence model 211 corresponding to a GPT model for providing a comprehensive solution that is versatile or public, not just for a specific organization or institution.

[0045] The electronic device 100 is capable of communicating with a second artificial intelligence model 211 stored in an external electronic device 200 for performing question-and-answering.

[0046] The second artificial intelligence model 211 may also include various engines for natural language processing and dialogue progression (e.g., DNN / RNN Engine, NLG, NLU, Dialog Engine, morphological analysis engine, individual analysis engine, semantic analysis engine, speech line analysis engine, related / similar word analysis engine, user-defined dictionary, etc.), generative models, etc., and may correspond to a chatbot or GPT model.

[0047] In this way, the electronic device 100, which can interface with both the internal first artificial intelligence model 111 and the external second artificial intelligence model 211, can provide responses to user questions through at least one of the first and second artificial intelligence models.

[0048] In relation to this, FIG. 2 is a flowchart illustrating the operation of an electronic device according to an embodiment of the present disclosure.

[0049] Referring to FIG. 2, the electronic device 100 may obtain a user's query for at least one document (S210).

[0050] The electronic device 100 may specify a document according to a user input to select at least one document already stored, or may provide a user interface (UI) for receiving document input, and may select and upload at least one document on the UI according to the user input.

[0051] The queries may include various questions or requests related to the document, where the queries may be simple questions, but may also include requests for specific tasks (e.g., data organization, schedule organization, document review, document excerpts, summaries, searches, etc.).

[0052] The question may be input in the form of voice or text and may be input in the form of a selection of at least one menu (eg, organize data, organize schedule, view document).

[0053] Meanwhile, the electronic device 100 may obtain a response to the question from the first artificial intelligence model 111 (S220). The first artificial intelligence model 111 may provide the response in the form of text or voice.

[0054] Specifically, the electronic device 100 may classify a document into a category (related to a question) based on a semantic analysis of the document by the first artificial intelligence model 111. In this case, the electronic device 100 may define the relevance between each keyword and a category based on a vector into which each keyword included in the document is converted, and may select a document category based on the relevance between each keyword and a category.

[0055] The categories may be set in various ways depending on the purpose or function of the electronic device 100 to the first artificial intelligence model 111. Specifically, the categories may be set in accordance with the nature of an organization, institution, etc. in which the electronic device 100 to the first artificial intelligence model 111 are utilized. For example, the categories may be divided into various categories such as official documents, private documents, reports, articles, papers, books, general documents, security documents, etc.

[0056] Alternatively, categories may be further divided within a particular field (e.g., the public field). For example, categories may include, but are not limited to, taxes, education administration, public safety, health care, real estate, road traffic, vulnerable population monitoring, and industrial employment.

[0057] The electronic device 100 may store documents in a knowledge base associated with the electronic device 100 based on the categorized categories, i.e., documents matching each category. The knowledge base may be included in the electronic device 100 and may be included in at least one database in communication with the electronic device 100.

[0058] In one embodiment, the electronic device 100 may obtain a response from the first artificial intelligence model 111 to the user's question based on at least one target document included in the same category as the document related to the input question. For example, when a question is input together with a document that falls under the education administration category, the electronic device 100 may provide a response using at least one target document related to the question among a plurality of documents belonging to the education administration category.

[0059] Specifically, the electronic device 100 may identify at least one keyword included in the user's query and identify a plurality of documents included in the same category as the document in the knowledge base, where the electronic device 100 may select at least one target document related to the identified keyword from among the plurality of documents, and obtain a response of the first artificial intelligence model 111 to the user's query based on the selected target document.

[0060] Meanwhile, the electronic device 100 may provide a final response to the response of the first artificial intelligence model 111, but may also provide a response by utilizing both the first artificial intelligence model 111 and the second artificial intelligence model 211.

[0061] Specifically, the electronic device 100 may update the question based on the response of the first artificial intelligence model 111 to the question (S230). Then, the electronic device 100 may obtain a response of the second artificial intelligence model 211 based on the updated question (S240).

[0062] In relation to this, FIG. 3 is a block diagram illustrating a functional configuration of an electronic device that uses a stored artificial intelligence model and an external artificial intelligence model according to one embodiment of the present disclosure.

[0063] 3, the electronic device 100 may include various engines constituting a first artificial intelligence model 111. Specifically, the first artificial intelligence model 111 may include, but is not limited to, a DNN / RNN engine, a natural language generation engine (NLG), a natural language understanding engine (NLU), a natural language processing engine (NLP), a dialogue engine, and the like.

[0064] 3, the electronic device 100 may include various engines / modules for management of the first artificial intelligence model 111. Specifically, the electronic device 100 may perform functions such as chatbot management for managing a chatbot specified by the first artificial intelligence model 111, log management, learning history management, sent / received dialogue history management, analysis / statistics / report generation, backup / restore, test management, monitoring, batch testing, etc.

[0065] The electronic device 100 may interface with a knowledge base in which documents are stored for each of various categories (e.g., categories 1-8), and the knowledge base may be utilized in the learning process and / or response generation process of the first artificial intelligence model 111.

[0066] In particular, the first AI model 111 is trained based on documents uploaded in response to user input and dialogue content with the user, and can respond to various questions and perform various tasks. Specifically, the electronic device 100 can train each engine constituting the first AI model 111 based on various documents uploaded.

[0067] 3, the electronic device 100 may acquire a user's question through a UI provided in the form of communication with a chatbot. For example, when a question is input as a voice (audio signal), the first artificial intelligence model 111 may perform natural language understanding of the voice-converted text to generate a response.

[0068] In this case, the electronic device 100 may further update the user's question based on the response generated by the first artificial intelligence model 111. For example, assume that a question of "How should the budget be allocated by part of education administration?" is acquired. In this case, the first artificial intelligence model 111 may collect related information based on documents in a related category (e.g., education administration) in the knowledge base. As a result, the first artificial intelligence model 111 may provide a response such as "Education administration is divided into national and local positions, and the total budget is OOO won" based on the contents of the documents in the knowledge base. Furthermore, the first artificial intelligence model 111 may provide information on the budget allocation history for each past year as an additional response.

[0069] At this time, the electronic device 100 can update the question by adding the response of the first artificial intelligence model 111 to the existing question, and can send the updated question to the external electronic device 200 to obtain the response of the second artificial intelligence model 211.

[0070] Meanwhile, in the process of updating the question, the electronic device 100 may replace personal information or sensitive information included in at least one of the user's question and the response of the first artificial intelligence model 111 with exemplary information. Personal information refers to various information related to personal identification information, passwords, privacy, etc., and may include, but is not limited to, a name, a resident ID number, a work history, a medical history, etc. Sensitive information may refer to information that has been given confidentiality or should not be disclosed within an organization or institution where the electronic device 100 is operated, and may correspond to confidential information that requires security, undisclosed information, information that is disclosed only to specific people or in specific situations, etc.

[0071] Meanwhile, a security grade may be set for at least one document related to a question. For example, assume that a security grade of a certain grade or higher is set for at least one of the document that is the subject of a user's question and / or the document used in the response process of the first artificial intelligence model 111. In this case, the electronic device 100 may identify security information included in the response of the first artificial intelligence model, and may replace the security information included in at least one of the user's question and the response of the first artificial intelligence model 111 with exemplary information in the process of updating the question.

[0072] As a result, it is possible to prevent a situation in which personal information or security information is leaked to the external electronic device 200 that does not only handle internal information but also has general-purpose or public uses.

[0073] Meanwhile, the electronic device 100 according to an embodiment of the present disclosure can selectively utilize the first artificial intelligence model 111 and the second artificial intelligence model 211 depending on the type of task requested in response to a user's question.

[0074] In relation to this, FIG. 4 is an algorithm for explaining an operation of an electronic device according to an embodiment of the present disclosure selectively utilizing an artificial intelligence model depending on the type of requested task.

[0075] Referring to FIG. 4, the electronic device 100 may obtain a user query for at least one document (S410) and identify a type of requested work according to the query (S420).

[0076] The types of tasks may be classified into, for example, a first task requiring both the first AI model 111 and the second AI model 211, and a second task that can be solved by the first AI model 111 alone. Here, the classification between the first task and the second task may be set in various ways. For example, the first task may include external search, response to specialized knowledge, response to common sense, coding, data organization, document generation, etc., and the second task may include, but is not limited to, internal search, response to internal knowledge, schedule organization, etc.

[0077] If the requested task is included in the pre-defined first task (S420-Y), the electronic device 100 may obtain a response from the first artificial intelligence model 111 to the question (S430). At this time, the electronic device 100 may update the question according to the response from the first artificial intelligence model 111 (S440), obtain a response from the second artificial intelligence model 211 to the updated question (S450), and finally provide a response (S460).

[0078] On the other hand, if the requested task is not included in the pre-defined first task (S420-N), the electronic device 100 obtains the response of the first artificial intelligence model 111 to the question (S430') and can immediately provide it as the final response (S460).

[0079] Meanwhile, in one embodiment, after the identified task is included in the second task and the response of the first artificial intelligence model is provided as the first response (e.g., S430' to S460), a user input requesting an additional response may be received. For example, an additional response may be requested as a result of the content of the response being insufficient or not matching the question.

[0080] In this case, the electronic device 100 may execute the above-described steps S430 to S460. Specifically, the electronic device 100 may update a question based on a response of a first artificial intelligence model, obtain a response of a second artificial intelligence model based on the updated question, and provide the response as a second response.

[0081] When the second response is provided in this manner, the electronic device 100 can identify whether there is additional information contained in the second response that is not contained in the first response, i.e., it is determined that the first artificial intelligence model was not derived, but there is additional information that the second artificial intelligence model of the external electronic device provides.

[0082] Here, if additional information exists, the electronic device 100 may update the first and second tasks so that the previously identified task is included in the first task and not the second task. If the same task is subsequently requested, the task is identified as the first task, and the electronic device 100 obtains a response through steps S430 to S450 rather than step S430'. As a result, more information desired by the user may be provided for the task in the future.

[0083] On the other hand, if there is no additional information, the electronic device 100 may train the first artificial intelligence model based on the second response. Specifically, the electronic device 100 may acquire a user input regarding the satisfaction of the user who confirmed the second response, and if a user evaluation that the second response is satisfied or that the second response is more appropriate than the first response is acquired, the electronic device 100 may train the first artificial intelligence model based on the second response. As a result, the first artificial intelligence model may be trained to provide a more appropriate response even if a request related to the same task is received in the future.

[0084] In a further embodiment, when the second response is provided as described above, the electronic device 100 may identify whether there is a contradiction between the second response and the first response. That is, it is confirmed whether the response of the first AI model and the response of the second AI model are logically inconsistent or conflict with each other. To this end, the electronic device 100 may semantically analyze the text constituting the first response and the second response to identify subjects, objects, verbs, negations, etc., and may identify whether there is a contradiction between the responses based on the semantic analysis.

[0085] For example, two responses may be identified as contradictory if they contain the same subject and object but the first response contains an affirmative form of a verb and the second response contains a negative form of the same verb.

[0086] At this time, the electronic device 100 may provide a message informing the user that the second response and the first response are contradictory to each other, and may receive user input for selecting an appropriate response between the first response and the second response.

[0087] Here, when the first response is selected, the electronic device 100 may identify documents stored in the knowledge base that match the text of the first response, i.e., documents that served as the basis for the first artificial intelligence model 111 to provide the first response. In this case, the electronic device 100 may inquire of the user whether the identified document has been made public.

[0088] If a user input is received to publish the identified document externally, the electronic device 100 may transmit the document to an external electronic device in which the second artificial intelligence model 211 is stored.

[0089] On the other hand, when a user input for not disclosing the identified document to the outside is received, the electronic device 100 may change at least one of the first tasks described above, which has a degree of relevance to the identified document of a certain value or more, to a second task. In this manner, the first task having a high degree of relevance to the document that cannot be disclosed is changed to the second task, thereby preventing confidential information from leaking to an external electronic device and establishing a response providing system that is optimized only for the user.

[0090] Here, the relevance level refers to the degree to which a corresponding document is related to a task, and the more frequently the corresponding document is used in the task, the higher the relevance level may be calculated. For example, in the process of providing a response to each first task requested in response to a question from one or more users, the electronic device 100 may record the frequency at which information included in each document is used, and the relevance level of each document to each first task may be calculated according to the recorded frequency.

[0091] On the other hand, if the second response described above is selected as the appropriate response, the electronic device 100 may similarly identify documents stored in the knowledge base that match the text of the first response, i.e., documents on which the first artificial intelligence model 111 provided the first response. In this case, the electronic device 100 may modify at least one piece of text related to the first response in the identified document according to the second response, or delete at least one piece of text related to the first response.

[0092] Meanwhile, FIGS. 5A-5B are diagrams illustrating how an electronic device according to an embodiment of the present disclosure may receive text input along with a user's question and provide one or more responses to the question.

[0093] The screens shown in Figures 5A and 5B may be UIs provided via a user terminal (e.g., a smartphone, a tablet PC, a desktop PC, a notebook PC, etc.) in cooperation with a user terminal via at least one web page or application by the electronic device 100 realized in a server. Alternatively, the screens shown in Figures 5A and 5B may be UIs displayed on a display of the electronic device 100 realized in a user terminal.

[0094] Referring to FIG. 5A, the electronic device 100 can obtain at least one document according to user input received via the UI 510.

[0095] For example, when the electronic device 100 is a server, the electronic device 100 can provide the UI 510 through a user terminal, and when a document (e.g., "ABCD related materials") stored in the user terminal is selected on the UI 510, the corresponding document can be received from the user terminal. However, it is of course possible for the electronic device 100 to be realized as a user terminal itself.

[0096] At this time, the electronic device 100 can train the first artificial intelligence model 111 based on the selected document and obtain user questions related to the document.

[0097] With reference to FIG. 5A in this regard, the electronic device 100 may provide a menu UI 521, a question and answer UI 522, a document management UI 523, and the like.

[0098] The menu UI 521 includes various menu items that the electronic device 100 can provide via the first artificial intelligence model 111, such as chatbot list, chatbot settings, chatbot creation, document analysis, MRC, and segment analysis.

[0099] The chatbot list is a menu item for providing a list of one or more chatbots provided via the electronic device 100. In this regard, the electronic device 100 may include a plurality of first artificial intelligence models classified according to functions or uses, and each of the first artificial intelligence models is provided in the form of a different chatbot. In this case, each of the first artificial intelligence models may learn in conjunction with a knowledge base of a different category, and a list of a plurality of chatbots may be provided, and at least one chatbot may be selected as a communication target. Alternatively, a plurality of chatbots may each provide a response to the same question.

[0100] The chatbot setting is a menu item for supporting various settings related to the chatbot, such as, but not limited to, the tasks performed by each chatbot, whether each chatbot is activated, etc.

[0101] The Create Chatbot menu item is for creating at least one chatbot. For example, at least one document for newly creating and training at least one chatbot can be input, and a chatbot trained on a new category based on the input document can be automatically generated on the electronic device 100.

[0102] Document Analysis is a menu item for performing analysis of an input document. The analysis can include various types of analysis, such as generating statistical information, extracting keywords, summarizing, and reliability analysis.

[0103] MRC (Machine Reading Comprehension) is a menu item for natural language processing and machine learning. Specifically, it can provide information on the learning and utilization of each engine (e.g., context understanding, dialogue generation, question and answer, etc.) that constitutes the first artificial intelligence model 111.

[0104] Syntax analysis is a menu item for analyzing the text constituting a document by syntax. For example, but not limited to, a UI may be provided in which the text is divided by syntax and displayed, and main keywords or related additional information may be provided for each division.

[0105] The question UI 522 is a UI for acquiring a question from the user. A question from the user can be input via the question UI 522, and the input question can be displayed.

[0106] The document management UI 523 is a UI for displaying documents related to a question. The document management UI 523 may include at least one document that is used in the process of the first artificial intelligence model 111 preparing a response to the question, as well as the document that is the subject of the question.

[0107] 5B, it is assumed that a question such as "Can you please review the service business bidding notice and organize it as a PDF file" is input via the question UI 522. In this case, the electronic device 100 can provide a plurality of responses 524-1, 2, 3 through a plurality of first AI models, respectively. At this time, each of the first AI models can provide a response by performing a different task (e.g., information browsing, file creation and provision, summary).

[0108] At this time, the documents used in the process of generating a response by each first artificial intelligence model can be displayed on the document management UI 523.

[0109] Meanwhile, FIG. 6 is a block diagram for explaining a configuration of an electronic device according to various embodiments of the present disclosure.

[0110] 6, the processor 130 of the electronic device 100 may include various functional modules, such as a knowledge management module 131, a model learning module 132, a response providing module 133, and a security management module 134. Each module may be implemented in software and / or hardware.

[0111] The knowledge management module 131 is a module for managing various documents by category to support the learning process, response generation process, etc. of the first artificial intelligence model 111.

[0112] The model learning module 132 is a module for training the first artificial intelligence model 111 based on categorized documents stored in the knowledge base.

[0113] The response providing module 133 is a module for providing a response through natural language understanding, natural language processing, etc. of the first artificial intelligence model 111. The response providing module 133 may provide various information corresponding to a response based on at least one document related to a user's question, and in this case, a search or analysis based on the natural language understanding of the artificial intelligence model 111 may be utilized.

[0114] The security management module 134 is a module for preventing leakage of security documents, security information, personal information, etc., to which a security grade is set. The security management module 134 can manage the security grade of each document included in the knowledge base, and the security grade can be set for each document according to a user input. Alternatively, the security management module 134 can automatically set the security grade of each document based on keywords included in each document. In this case, the security management module 134 can identify whether each keyword corresponds to personal information or sensitive information based on a semantic analysis of the first artificial intelligence model 111. Here, the more keywords that correspond to personal information or sensitive information, the higher the security grade of the document can be set.

[0115] Referring to FIG. 6, the electronic device 100 may further include a user input unit 140, an output unit 150, etc., in addition to a memory 110, a communication unit 120, and a processor .

[0116] The user input unit 140 is configured to receive various commands or information input from a user. The user input unit 140 may be realized by at least one button, a touch pad, a touch screen, a microphone, a camera, a sensor, etc. The electronic device 100 may also be connected to various user input devices (e.g., a keyboard, a mouse, etc.) having at least one keypad, button, motion sensor, etc.

[0117] The output unit 150 is configured to output various information, and may include a display, a speaker, an earphone / headset terminal, etc. In one embodiment, the UI screens of Fig. 5A and Fig. 5B may be provided via a display of the output unit 150.

[0118] On the other hand, the various embodiments described above may be realized in such a manner that two or more embodiments are combined with each other, unless they conflict or contradict each other.

[0119] Meanwhile, the various embodiments described above may be realized in a recording medium readable by a computer or a similar device using software, hardware, or a combination thereof.

[0120] According to a hardware implementation, the embodiments described in this disclosure may be implemented using at least one of Application Specific Integrated Circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, micro-controllers, microprocessors, and other electrical units for performing functions.

[0121] In some cases, the embodiments described herein may be implemented in the processor itself. According to a software implementation, the embodiments, such as the procedures and functions described herein, may be implemented in separate software modules. Each of such software modules may perform one or more of the functions and operations described herein.

[0122] Meanwhile, computer instructions or computer programs for executing processing operations in an electronic device such as a robot according to the various embodiments of the present disclosure may be stored in a non-transitory computer-readable medium, and when executed by a processor of the specific device, the computer instructions or computer programs cause the specific device to execute the processing operations in the electronic device according to the various embodiments.

[0123] A non-transitory computer-readable medium refers to a medium that can store data semi-permanently and can be read by a device, rather than a medium that stores data for a short moment, such as a register, cache, memory, etc. Examples of non-transitory computer-readable media include CDs, DVDs, hard disks, Blu-ray disks, USBs, memory cards, ROMs, etc.

[0124] Although preferred embodiments of the present disclosure have been shown and described above, the present disclosure is not limited to the specific embodiments described above, and it goes without saying that various modifications can be made by a person having ordinary skill in the art to which the disclosure pertains without departing from the gist of the present disclosure as claimed in the claims, and such modifications should not be understood separately from the technical ideas or perspectives of the present disclosure. [Explanation of symbols]

[0125] 100: Electronic devices

Claims

1. In an electronic device, a memory containing a first artificial intelligence model for performing question-and-answer processing; a communication unit for communicating with an external electronic device including a second artificial intelligence model for performing a question-and-answer process; a processor for providing a response to a user's query via at least one of the first artificial intelligence model and the second artificial intelligence model; The processor, obtaining a query of the user for at least one document, and obtaining a response of the first artificial intelligence model to the query; updating the query based on a response of the first artificial intelligence model; Obtaining a response from the second artificial intelligence model based on the updated question; The processor, classifying a category of the document based on a semantic analysis of the first artificial intelligence model for the document; storing the document in a knowledge base associated with the electronic device based on the classified category; The processor, An electronic device that obtains a response of the first artificial intelligence model to the user's query based on at least one target document included in the same category as the document in the knowledge base.

2. The processor, Identifying at least one keyword contained in the user's query; identifying a plurality of documents in the knowledge base that are in the same category as the document; selecting at least one target document related to the keyword from among the plurality of documents; The electronic device of claim 1 , further comprising: a processor configured to generate a response of the first artificial intelligence model to the user's query based on the selected target document.

3. The processor, The electronic device of claim 1 , wherein in the process of updating the query, personal information included in at least one of the user query and the response of the first artificial intelligence model is replaced with example information.

4. The processor, When a security level of a certain level or higher is set for at least one of a first document related to the question and a second document related to the response of the first artificial intelligence model, identifying security information included in the response of the first artificial intelligence model; The electronic device of claim 1 , wherein in the process of updating the query, the security information included in at least one of the user query and the first artificial intelligence model is replaced with example information.

5. The artificial intelligence model is At least one of an engine including a recurrent neural network (RNN) model, a natural language generation engine, a natural language processing engine, a natural language understanding engine, and a dialogue engine; The processor, The electronic device of claim 1 , further comprising: a processor for processing said at least one engine that configures said artificial intelligence model based on said documents.

6. The processor, identifying at least one requested task pursuant to said user query; If the identified task is included in a first task that has been set up, obtain a response from the first artificial intelligence model to the question, update the question based on the response from the first artificial intelligence model, and obtain and provide a response from the second artificial intelligence model based on the updated question; The electronic device of claim 1 , further comprising: if the identified task is included in a second predefined task, the electronic device obtains and provides a response of the first artificial intelligence model to the query.

7. The processor, When a user input requesting an additional response is received after the identified task is included in the second task and a response of the first artificial intelligence model is provided as a first response, the question is updated based on the response of the first artificial intelligence model, and a response of the second artificial intelligence model is obtained based on the updated question and provided as a second response; If the second response is provided, identifying whether there is additional information included in the second response that is not included in the first response; If the additional information exists, updating the first service and the second service such that the identified service is included in the first service but not in the second service; The electronic device of claim 6 , further comprising: training the first artificial intelligence model based on the second response if the additional information is not present.

8. 1. A method of controlling an electronic device including a first artificial intelligence model for performing question-and-answer processing, comprising the steps of: obtaining a user query for at least one document, the electronic device obtaining a response of the first artificial intelligence model to the query; updating the query based on a response of the first artificial intelligence model by the electronic device; and acquiring a response of a second artificial intelligence model that performs question-answering and is stored in the external electronic device based on the updated question by the electronic device; The method for controlling an electronic device includes: classifying a category of the document based on a semantic analysis of the first artificial intelligence model for the document; storing the document in a knowledge base associated with the electronic device based on the classified category; The step of obtaining a response of the first artificial intelligence model includes: A method for controlling an electronic device, comprising obtaining a response of the first artificial intelligence model to the user's query based on at least one target document included in the same category as the document in the knowledge base.

Citation Information

Patent Citations

  • Apparatus for adding data for chatbot based on machine learning

    KR1020190055425A

  • electronic apparatus for responding to question using multi chat-bot and control method thereof

    KR1020220046964A

  • method and apparatus for automatically generating training dataset for FAQ and chatbot based on natural language processing using deep learning

    KR102436549B1