CN116508016A

CN116508016A - Electronic device for determining time of maintaining conversation of chat robot and operation method thereof

Info

Publication number: CN116508016A
Application number: CN202280007673.2A
Authority: CN
Inventors: 金湖泳; 杨东一; 安在炯; 李知衍; 郑哲丞
Original assignee: Samsung Electronics Co Ltd
Current assignee: Samsung Electronics Co Ltd
Priority date: 2021-01-29
Filing date: 2022-01-21
Publication date: 2023-07-28
Also published as: US11700311B2; US20220247822A1; EP4207712A1; EP4207712A4

Abstract

An electronic device for determining a duration of a session according to characteristics of a response message in a chat robot and an operating method thereof are provided. The electronic device is configured to: determining a default session time based on the difficulty level of the response message; determining an additional session time based on session history information through the chat robot before outputting the response message; determining a session duration, the session duration being a sum of a default session time and an additional session time; and providing the session duration.

Description

Electronic device for determining time of maintaining conversation of chat robot and operation method thereof

Technical Field

The present disclosure relates to an electronic device for determining a length of time a chat robot maintains a session in which a response message is provided for a user input message, and an operating method thereof.

Background

When a user inputs a query message in an electronic device or asks a question through a voice message, a system for providing a response to the user's question is used. More specifically, recently, with the development of artificial intelligence and big data technology, intelligent response systems such as chat robots are widely used. A "chat robot" is an abbreviation of a chat robot and is an artificial intelligence service configured to perform operations that provide information about a problem, for example, by conversational with a user using voice signals or text, or provide a service for a request. As a service for interaction through a messenger application, a chat robot is provided through preset rules or artificial intelligence techniques.

In a session by a chat robot, a session duration is determined according to personal information protection of a user, and the session ends when the session duration elapses. The session duration is updated by user input. When the session ends before the problem is completely solved as the session duration passes, in order to return to the previously ongoing problem solving scenario, the user may have to input the message input at the previous point in time again and be provided with the same response message again, which can be cumbersome and time consuming. Further, even though different processing times are required according to the problem solving scenario, since the total session duration is provided, there is a problem in that the session is accidentally terminated due to the time taken for the user to solve the problem.

Extending the session duration of all problem-solving scenarios may be considered a solution. However, this solution results in additional resources to be used and a higher risk of personal information leakage due to an increased number of simultaneous management sessions and a corresponding increase in costs.

The above information is presented merely as background information to aid in the understanding of the present disclosure. No determination has been made nor is any assertion regarding whether any of the above may be suitable as prior art with respect to the present disclosure.

Disclosure of Invention

[ technical problem ]

Aspects of the present disclosure are directed to solving at least the problems and/or disadvantages described above and to providing at least the advantages described below. Accordingly, it is an aspect of the present disclosure to provide an electronic device for providing a session duration suitable for a problem solving scenario by determining a default session time and an additional session time according to characteristics of a reply message.

Additional aspects will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the presented embodiments.

[ solution ]

According to one aspect of the present disclosure, a method of determining a session duration of a chat robot is provided. The method comprises the following steps: outputting a response message for the input message of the user; determining a default session time for which the session is maintained based on the difficulty level of the outputted response message; determining an additional session time based on session history information through the chat robot before outputting the response message; waiting for the user to enter an additional message during a session duration, the session duration being a sum of a default session time and an additional session time; and determining an end of the session based on whether the additional message is entered.

According to an embodiment of the present disclosure, determining a default session time includes: obtaining a marking value of the difficulty level by applying the response message as input to the deep neural network model; and determining the session time mapped to correspond to the obtained flag value as a default session time.

According to an embodiment of the present disclosure, a deep neural network model is trained by supervised learning, wherein a plurality of training messages are applied as inputs and a labeled value of a difficulty level is applied as an output true value (groudtluth).

According to embodiments of the present disclosure, the dialog history information may include information related to at least one of: the number of response messages previously output in a session before the response message is output, or the session use time in a scene performed according to the previously output response message.

According to embodiments of the present disclosure, the additional session time may be determined by a preset ratio of the sum of session use times for each scene.

According to an embodiment of the present disclosure, determining the additional session time may include: a conversation time usage ratio is calculated by performing an operation of dividing a conversation usage time for each scene by a default conversation time, the calculated conversation time usage ratio is compared with a preset threshold, and an additional conversation time is determined based on the comparison result.

According to an embodiment of the present disclosure, determining the additional session time may include: an average value of session use times before outputting the response message is calculated, the calculated average value is compared with a preset threshold value, and an additional session time is determined based on the comparison result.

According to embodiments of the present disclosure, determining the session end may include: a session end notification is output that provides information about the end of the session before the session duration has elapsed.

According to an embodiment of the present disclosure, determining the session duration may further include: storing session use history information of a plurality of users with respect to the outputted response message; and adjusting a default session time based on the stored session usage history information for the plurality of users.

According to embodiments of the present disclosure, the session usage history information may include statistics of a plurality of users regarding at least one of: the number of session extension requests at a point of time before outputting the response message, the number of scenes executed before the response message, or the session use time for the response message.

According to another aspect of the present disclosure, an electronic device for determining a session duration of a chat robot is provided. The electronic device includes a communication interface configured to perform transmission and reception of data with other devices, a memory, and a processor configured to execute one or more instructions. Wherein the processor is further configured to execute one or more instructions to: obtaining an input message input by a user through a communication interface; outputting a response message for the input message of the user; determining a default session time for which the session is maintained based on the difficulty level of the outputted response message; determining an additional session time based on session history information through the chat robot before outputting the response message; waiting for the user to enter an additional message during a session duration, the session duration being a sum of a default session time and an additional session time; and determining an end of the session based on whether the additional message is entered.

According to embodiments of the present disclosure, the processor may obtain an output tag value of the difficulty level by applying the response message as an input to the deep neural network model, and determine a session time mapped to correspond to the obtained tag value as a default session time.

According to embodiments of the present disclosure, the deep neural network model may be an artificial intelligence model trained by supervised learning, where a plurality of training messages are applied as inputs and a labeled value of difficulty level is applied as an output truth value.

According to embodiments of the present disclosure, the processor may be further configured to execute one or more instructions to: the additional session time is determined by a preset ratio of the sum of session usage times for each scene.

According to embodiments of the present disclosure, the processor may be further configured to execute one or more instructions to: a conversation time usage ratio is calculated by performing an operation of dividing a conversation usage time of each scene by a default conversation time, dividing the conversation usage time of each scene by the default conversation time, comparing the calculated conversation time usage ratio with a preset threshold, and determining an additional conversation time based on the comparison result.

According to embodiments of the present disclosure, the processor may be further configured to execute one or more instructions to: a session end notification is output that provides information about the end of the session before the session duration has elapsed.

According to embodiments of the present disclosure, the processor may be further configured to execute one or more instructions to: session use history information of a plurality of users regarding the outputted response message is stored in a database of the memory, and a default session time is adjusted based on the stored session use history information of the plurality of users.

According to another aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer-readable storage medium having a program stored therein to be executed on a computer.

Drawings

The foregoing and other aspects, features, and advantages of certain embodiments of the present disclosure will become more apparent from the following description, taken in conjunction with the accompanying drawings, in which:

fig. 1 is a conceptual diagram illustrating a method of providing a chat bot service to a client device over a network performed by an electronic device according to an embodiment of the disclosure;

fig. 2 is a conceptual diagram illustrating a method of determining a session duration performed by an electronic device according to an embodiment of the present disclosure;

FIG. 3 is a block diagram of components of an electronic device according to an embodiment of the present disclosure;

FIG. 4 is a flowchart of a method of operation of an electronic device according to an embodiment of the present disclosure;

FIG. 5a is a diagram illustrating a method performed by an electronic device for training a deep neural network model for classifying a difficulty level of a response message, according to an embodiment of the present disclosure;

FIG. 5b is a diagram illustrating a method performed by an electronic device to obtain information about a difficulty level of a response message using a deep neural network model, according to an embodiment of the present disclosure;

FIG. 5c is a diagram illustrating a method performed by an electronic device to determine a default session time based on a difficulty level in accordance with an embodiment of the present disclosure;

Fig. 6 is a flowchart illustrating a method of operation of an electronic device according to an embodiment of the present disclosure;

FIG. 7 is a schedule showing a relationship between a default session time and a session use time according to an input message input by a user and a response message provided by a chat bot, respectively, according to an embodiment of the present disclosure;

FIG. 8 is a flowchart illustrating a method performed by an electronic device to determine whether to provide additional session time, according to an embodiment of the present disclosure;

FIG. 9 is a flowchart illustrating a method performed by an electronic device to determine whether to provide additional session time, according to an embodiment of the present disclosure;

FIG. 10 is a flowchart illustrating a method performed by an electronic device to determine whether to provide additional session time, according to an embodiment of the present disclosure;

FIG. 11 is a flowchart illustrating a method of outputting a session end notification performed by an electronic device according to an embodiment of the present disclosure;

FIG. 12 is a flowchart illustrating a method performed by an electronic device to adjust a default session time, according to an embodiment of the present disclosure; and

fig. 13 is a flowchart illustrating a method of extending or shortening a default session time performed by an electronic device according to an embodiment of the present disclosure.

Throughout the drawings, it should be noted that the same reference numerals are used to describe the same or similar elements, features and structures.

Detailed Description

The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of various embodiments of the disclosure as defined by the claims and their equivalents. It includes various specific details to aid understanding, but these are to be considered exemplary only. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the various embodiments described herein can be made without departing from the scope and spirit of the present disclosure. In addition, descriptions of well-known functions and constructions may be omitted for clarity and conciseness.

The terms and words used in the following description and claims are not limited to bookend meanings, but are used only by the inventors to enable clear and consistent understanding of the present disclosure. Accordingly, it should be apparent to those skilled in the art that the following descriptions of the various embodiments of the present disclosure are provided for illustration only and not for the purpose of limiting the disclosure as defined by the appended claims and their equivalents.

It is to be understood that the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a component surface" includes reference to one or more such surfaces.

Throughout this disclosure, the expression "at least one of a, b, or c" means a only, b only, c only, both a and b, both a and c, both b and c, all a, b, and c, or variants thereof.

Although terms used in the embodiments of the present specification are selected from general terms currently in common use in consideration of functions in the present disclosure, these terms may be changed according to the intention of one of ordinary skill in the art, judicial precedents, or introduction of new technologies. Furthermore, in certain cases, the applicant may actively choose a term, and in such cases the meaning of the term is disclosed in the respective description section of the present disclosure. Accordingly, the terms used in the specification should not be defined by simple names of the terms, but should be defined by meanings of the terms and throughout the present disclosure.

It is to be understood that the singular form of a noun corresponding to an entry may include one or more things unless the context clearly dictates otherwise. All terms (including technical or scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art described in the specification.

Throughout this disclosure, when a portion is assumed to include a certain component, the term "comprising" means that the corresponding component may further include other components, unless a specific meaning is written contrary to the corresponding component. As used in this specification, terms such as "unit" or "module" mean a unit for processing at least one function or operation, and may be implemented in hardware, software, or a combination of hardware and software.

The expression "configured (or arranged)" as used in the present specification may be replaced with, for example, "adapted to", "having the capability of … …", "designed to", "adapted to", "made of" or "capable" as the case may be. The term "configured (or arranged)" does not always mean "specifically designed to be implemented by hardware only". Alternatively, in some cases, the expression "a device configured as … …" may mean that the device "may operate with another device or component. For example, the phrase "a processor configured (or arranged) to perform A, B and C" may be a special purpose processor (e.g., an embedded processor) for performing the respective operations or a general purpose processor (e.g., a Central Processing Unit (CPU) or an application processor) that may perform the respective operations by executing at least one software program stored in a memory.

In the present disclosure, "chat robot" is an abbreviation of chat robot and is an artificial intelligence service configured to perform operations that provide information about a problem, for example, by having a conversation with a user using a voice signal or text, or provide a service for a request. As a service for interaction through a messenger application, a chat robot is provided through preset rules or artificial intelligence techniques.

In this disclosure, a "conversation" is a group of user messages and chat robot response messages entered by the chat robot during a period of time.

In this disclosure, a "session duration" is the time that a session is maintained without termination. The session duration starts from the time when a response message of the chat bot to the user input message is output. The session duration is updated and initialized as the user enters a message.

In the present disclosure, the "session use time" is a time when a user uses a session, and refers to a time interval between when a response message of a chat robot is output and when the user inputs a response message or an additional message for the response message.

In this disclosure, a "scenario" is a problem-solving operation (or process) provided by a chat bot for an input message of a user, such as requesting, querying, replying to, or providing information through a response message. For example, when the input message is "application regarding failure", the scenario may refer to an operation of solving a problem by the chat robot through a response message requesting, asking, replying to, or providing information until the application regarding failure is completed. Session use time and scenario will be described with reference to fig. 7.

Fig. 1 is a conceptual diagram illustrating a method of providing a chat bot service to a client device over a network performed by an electronic device according to an embodiment of the disclosure.

Referring to fig. 1, the electronic device 1000 is a computing device that receives input messages from users of client devices 3001, 3002, and 3003 and provides response messages for the input messages. According to embodiments of the present disclosure, the electronic device 1000 may be implemented as a chat bot that provides response messages to incoming messages, but is not limited thereto.

According to embodiments of the present disclosure, the electronic device 1000 may include a server or a workstation. However, the present disclosure is not limited thereto, and the electronic device 1000 may include at least one of a smart phone, a tablet Personal Computer (PC), a mobile phone, a video phone, an electronic book reader, a desktop Personal Computer (PC), a laptop PC, a netbook computer, a Personal Digital Assistant (PDA), or a Portable Multimedia Player (PMP), for example.

Electronic device 1000 may send data to client devices 3001, 3002, and 3003 or receive data from client devices 3001, 3002, and 3003 over network 2000. The network 2000 may be configured not to be limited to a certain communication method, such as wired or wireless communication. Network 2000 may include at least one wired or wireless data communication method including, for example, ethernet, wired or wireless Local Area Network (LAN), wi-Fi direct (WFD), and Wireless gigabit alliance (WiGig).

The client devices 3001, 3002, and 3003 are terminals used by the user to receive chat bot services from the electronic device 1000 through the chat bot application. According to embodiments of the present disclosure, a user may execute a chat bot application through the client devices 3001, 3002, and 3003, input a text message to the chat bot application, and receive a response message from the electronic device 1000. According to embodiments of the present disclosure, the client devices 3001, 3002, and 3003 may receive voice input from a user, convert the voice input into text and send the text to the electronic device 1000, and receive a response message to the text from the electronic device 1000.

Client devices 3001, 3002, and 3003 may include, for example, smart phones, tablet PCs, mobile phones, video phones, electronic book readers, desktop PCs, laptop PCs, web book computers, PDAs, or PMPs.

The electronic device 1000 of the present disclosure is not limited to receiving input messages from the client devices 3001, 3002, and 3003 through the network 2000 and providing response messages to the client devices 3001, 3002, and 3003, as shown in fig. 1. According to embodiments of the present disclosure, unlike the embodiment shown in fig. 1, the electronic device 1000 may receive an input message directly from a user without intervention of the network 2000 and output a response message to the input message.

Fig. 2 is a conceptual diagram illustrating a method of determining a session duration performed by an electronic device according to an embodiment of the present disclosure.

Referring to fig. 2, a chat robot conversation screen is shown. The electronic device 1000 may receive the first input message 210 "food in the refrigerator is not frozen" from the user and output the first response message 220 for the first input message 210. The first response message 220 may be a message requesting confirmation from the user for the first input message 210. For example, the first response message 220 may be "do you have checked that the power line of the refrigerator is plugged in and out? "

The electronic device 1000 may initialize a session time when outputting the first response message 220. The electronic device 1000 may wait for the user to enter additional messages during a preset default session duration. The user may use the session time due to checking the power line.

When the user inputs the second input message 230, the electronic device 1000 may output a second response message 240 for the second input message 230. According to embodiments of the present disclosure, the second input message 230 may include a user's response or additional questions to the first response message 220. In the embodiment of the present disclosure shown in fig. 2, the second input message 230 may be "yes", with the power cord plugged in. The second response message 240 may be a message for additional requests, responses, or information provided by the chat bot for the second input message 230. In the presently disclosed embodiment shown in fig. 2, the second response message 240 may be "do you have checked whether the cooling engine of the refrigerator is operating? "the electronic device 1000 may initialize the session time when outputting the second response message 240.

The electronic device 1000 may determine the default session time based on the difficulty level of the second response message 240. The difficulty level of the second response message 240 refers to the difficulty level of solving the problem by requesting, asking, or providing information in response to the message. According to embodiments of the present disclosure, the electronic device 1000 may perform training of classifying the response message according to the difficulty level using the deep neural network model, and obtain information about the difficulty level of the second response message 240 through the deep neural network model. The electronic device 1000 may determine the session time mapped to correspond to the difficulty level as a default session time.

The electronic device 1000 may wait for the user to enter additional messages during a default session time. When no additional message of the user is input even when the default session time elapses, the electronic device 1000 can determine whether to provide the additional session time. The electronic device 1000 may determine the additional session time based on the currently output response message (i.e., the dialog history information prior to outputting the second response message 240). According to embodiments of the present disclosure, the electronic device 1000 may determine the additional session time based on information related to at least one of: regarding the number of response messages previously output before the second response message 240 is output, or session use time used in a scene performed by the previously output response message.

In the embodiment of the present disclosure shown in fig. 2, the electronic device 1000 may recognize the first response message 220 output before outputting the current message (i.e., the second response message 240), obtain information on a session usage time used by the user during a scene of the first response message 220 until the user inputs the second input message 230, and determine an additional session time based on the number of the first response messages 220 and the session usage time in the previous scene.

The electronic device 1000 may determine a certain ratio of the default session time as the additional session time. For example, the electronic device 1000 may determine 20% of the default session time as the additional session time.

The electronic device 1000 may determine the sum of the default session time and the additional session time as the session duration. The electronic device 1000 may identify whether the session duration has elapsed and may output the session end notification 250 when the session duration has elapsed.

According to embodiments of the present disclosure, the electronic device 1000 may extend or shorten the default session time based on statistics regarding dialogue history information of other users for the current response message.

In a session by a chat robot, a session duration is determined according to personal information protection of a user, and the session is terminated when the session duration is terminated. In the chat robot according to the related art, since the session may be terminated before the problem is completely solved when the session duration elapses, in order to return to the previously ongoing problem solving scenario, the user may have to input the message input at the previous point in time again and be provided with the same response message again, which may be cumbersome and time-consuming. Further, even though different processing times are required for the chat robot according to the problem solving scenario according to the related art, since the total session duration is provided, there is a problem in that the session is accidentally terminated due to the time taken for the user to solve the problem. Extending the session duration of all problem-solving scenarios may be considered a solution. However, such a solution would result in an increased number of sessions being simultaneously managed by the chat robot, would correspondingly increase resource usage, and would increase network costs.

The electronic device 1000 according to the embodiment of the present disclosure may provide a chat robot service that adaptively controls a session time according to a problem solving scenario by: the method includes determining a default session time according to a difficulty level of a response message, determining whether to provide an additional session time based on dialog history information before a currently output response message, and extending or shortening the session time by using statistical data regarding session usage history information of a plurality of users. Accordingly, the electronic device 1000 of the present disclosure may have the following technical effects: preventing the user from consuming time and inconvenience due to the repetition of providing the same response message to the user as the session duration elapses, and improving resource usage and network cost efficiency.

Fig. 3 is a block diagram of components of an electronic device according to an embodiment of the present disclosure.

Referring to fig. 3, an electronic device 1000 may include a communication interface 1100, a processor 1200, and a memory 1300. Fig. 3 shows only basic components for explaining the functions and/or operations of the electronic device 1000, and components included in the electronic device 1000 are not limited to those shown in fig. 3.

The communication interface 1100 may be configured to perform transmission and reception of data between the electronic device 1000 and other devices (e.g., client devices 3001, 3002, and 3003). The communication interface 1100 may perform data communication with a server or other device by using at least one of wired and wireless data communication methods including ethernet, wired or wireless LAN, wi-Fi, WFD, or WiGig. According to embodiments of the present disclosure, the electronic device 1000 may include a server that receives text data regarding an input message from other devices (e.g., client devices 3001, 3002, and 3003) through the communication interface 1100 and transmits text data regarding a response message to the other devices.

Although not shown in fig. 3, the electronic device 1000 may include a speech input interface and an Automatic Speech Recognition (ASR) model. The voice input interface may include means for receiving voice input (e.g., user questions from a user). The voice input interface may include, for example, a microphone. The voice input interface may receive voice input (e.g., a user's utterance) from a user through a microphone and obtain a voice signal from the received voice input. According to embodiments of the present disclosure, the processor 1200 of the electronic device 1000 may receive a user's voice input through a microphone, convert the received voice input into an acoustic signal, and obtain the voice signal by removing noise (e.g., non-voice components) from the acoustic signal.

The ASR model is a speech recognition model that recognizes user speech and is a model trained to convert speech input received from a user into text and output the converted text. According to embodiments of the present disclosure, the electronic device 1000 may convert speech input into text by using an ASR model.

Although not shown in the drawings, the electronic device 1000 may include a voice preprocessing module having a function of detecting a specified voice input (e.g., a wake-up input such as "Hi Bixby", "Ok Google", etc.) or a function of preprocessing a voice signal obtained from some of the voice inputs.

Processor 1200 may be configured to control electronic device 1000 by reading and executing one or more instructions or program code to perform operations and/or functions. The memory 1300 may store instructions or program codes executed by the processor 1200, and the processor 1200 may execute instructions or program codes loaded from the memory 1300, but is not limited thereto. The processor 1200 itself may include instructions or program code.

Processor 1200 may include hardware components that perform arithmetic, logic, and input/output operations, as well as signal processing. The processor 1200 may include at least one of a Central Processing Unit (CPU), a microprocessor, a Graphics Processing Unit (GPU), an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), or a Field Programmable Gate Array (FPGA), but is not limited thereto.

Memory 1300 may store instructions and program codes readable by processor 1200. The memory 1300 may include at least one type of storage medium such as a flash memory type, a hard disk type, a multimedia card micro type, a card type memory such as a Secure Digital (SD) or extreme digital (xD) -photo card memory (xD memory), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a Programmable Read Only Memory (PROM), a magnetic memory, a magnetic disk, or an optical disk.

Processor 1200 may include a default session time determination module 1210, an additional session time determination module 1220, an end notification module 1230, and a session time adjustment module 1240.

The processor 1200 may receive input messages of a user through the communication interface 1100. The processor 1200 may identify a domain, an intention, and an entity of the input message by analyzing the received user input message, and output a response message for the identified domain, intention, and entity of the input message. According to embodiments of the present disclosure, the processor 1200 may parse text in units of morphemes, words, or phrases by using a natural language understanding model, and infer the meaning of the parsed text by using linguistic features (e.g., syntax elements) of the morphemes, words, or phrases. The processor 1200 can identify domains, intents, and entities by comparing the inferred word meanings to each of the predefined domains, intents, and entities provided by the natural language understanding model. According to embodiments of the present disclosure, the processor 1200 may output a set of response messages corresponding to a domain, an intention, and an entity.

The default session time determination module 1210 is a module configured to obtain information about the difficulty level 540 of the response message by analyzing the output response message, and determine a default session time based on the difficulty level 540 of the response message. According to embodiments of the present disclosure, the default session time determination module 1210 may include a deep neural network model 1214. However, the present disclosure is not limited thereto, and the deep neural network model 1214 may be executed by the processor 1200 or stored in the memory 1300 as a model independent of the default session time determination module 1210. The processor 1200 may apply the text data of the output response message to the deep neural network model 1214 as input and obtain the output marker value of the difficulty level 540 via the deep neural network model 1214 by executing instructions or program code with respect to the default session time determination module 1210.

The deep neural network model 1214 may be a model trained to classify response messages according to the difficulty level 540. The deep neural network model 1214 may be implemented as a trained neural model some time before the response message is output. According to embodiments of the present disclosure, the deep neural network model 1214 may be trained by supervised learning, wherein a plurality of training messages previously obtained are applied as inputs and the marker value of the difficulty level 540 set for each of the plurality of training messages is applied as an output true value. Here, "difficulty level" 540 refers to the difficulty level of solving a problem by a request, problem, response, or information provided by a message. The tag value of difficulty level 540 may be defined as an integer value, such as 0, 1, 2, … …, n, but is not limited thereto.

The deep neural network model 1214 may be implemented as, for example, a convolutional neural network model (CNN), but is not limited thereto. According to embodiments of the present disclosure, the deep neural network model may be implemented as a deep learning-based neural network model, such as a Recurrent Neural Network (RNN), a limited Boltzmann machine (RBM), a Deep Belief Network (DBN), a bi-directional recurrent deep neural network (BRDNN), or a deep Q network.

The processor 1200 may input a response message to the deep neural network model 1214 and obtain a marker value for the difficulty level 540 by using the deep neural network model 1214, by which the response message may be classified by training. According to an embodiment of the present disclosure, the processor 1200 may search a tag value for the obtained difficulty level 540 from a session time Database (DB) 1310, and obtain information on a session time mapped to correspond to the tag value from the session time DB 1310.

The session time DB 1310 is a database storing session times according to the difficulty level 540. According to an embodiment of the present disclosure, the session time DB 1310 may store a plurality of difficulty mark values and data of a plurality of session times. The plurality of session times may be mapped to a tag value corresponding to each of the plurality of difficulties. According to an embodiment of the present disclosure, the session time DB 1310 may store a tag value of each of a plurality of difficulties and data of a plurality of session times as a key-value (key-value) type.

Detailed embodiments of the present disclosure for determining a default session time for a response message by using the default session time determination module 1210 will be described with reference to fig. 5a to 5c and fig. 6.

The additional session time determination module 1220 is a module configured to: information is received from the default session time determination module 1210 regarding whether the default session time for the response message has elapsed and additional session times for the response message are determined. The additional session time determination module 1220 may determine the additional session time based on the dialog history information before outputting the response message. According to embodiments of the present disclosure, the dialog history information may include information related to at least one of: the number of response messages output in a previous session before outputting the response message, or a session use time for each scene according to the previously output response message.

The processor 1200 may determine whether to provide the additional session time based on the dialog history information by executing instructions or program code regarding the additional session time determination module 1220. According to an embodiment of the present disclosure, the processor 1200 may calculate a usage ratio of the session time by dividing the session usage time of each scene performed before outputting the response message by a default session time, and determine whether to provide the additional session time based on the calculated usage ratio of the session time. According to another embodiment of the present disclosure, the processor 1200 may calculate an average value of session use times for each scene performed before outputting the response message, and determine whether to provide additional session times based on the calculated average value.

The processor 1200 may determine a certain ratio of the sum of session usage times for each scene as the additional session time. Detailed embodiments of the present disclosure in which the processor 1200 determines additional session times will be described with reference to fig. 8 through 10.

The end notification module 1230 is a module configured to output a notification message regarding the end of the session when it is confirmed that the session duration has elapsed. The end notification module 1230 may include a natural language generation model according to an embodiment of the disclosure. The end notification module 1230 may generate a message informing the user that the session hold time has elapsed by using the intention and the entity associated with the response message, and output the generated message. The session end message may be generated and output as, for example, "is the problem with the refrigerator solved? The chat is ended. "

The session duration may be determined by the sum of the default session time and the additional session time. The processor 1200 may recognize whether the session duration has elapsed by executing instructions or program code regarding the end notification module 1230, and output a session end notification message according to the recognition result. An embodiment of the present disclosure regarding outputting a session end notification message will be described with reference to fig. 11.

The session time adjustment module 1240 is a module configured to adjust a default session time based on session usage history information of a plurality of users with respect to the outputted response message. The session time adjustment module 1240 may receive information about the passage of the session duration from the end notification module 1230, and thus may search the session use history statistics DB 1320 for statistics about session use history information related to a currently output response message. The session time adjustment module 1240 may obtain session use history information of a plurality of users according to the search results of the session use history statistics DB 1320, and may lengthen or shorten a default session time based on the session use history information. The session time adjustment module 1240 may provide information about the extended or reduced default session time to the default session time determination module 1210.

The session use history statistical data DB 1320 is a database for storing statistical data on session use histories of a plurality of users. According to embodiments of the present disclosure, the session usage history information may include statistics of a plurality of users, the statistics of the plurality of users being related to at least one of: the number of session extension requests with a point in time before outputting the current response message, the number of scenes performed before the response message, or the session use time for the response message.

The processor 1200 may search statistical data on session use histories of a plurality of users from the session use history statistical data DB 1320 by executing instructions or program codes on the session time adjustment module 1240 and obtain session use history information according to the search result. The processor 1200 may extend or shorten the default session time based on the session usage history information. Detailed embodiments of the present disclosure in which the processor 1200 adjusts the default session time based on the session use history information will be described with reference to fig. 12 and 13.

The session time DB 1310 and the session usage history statistics DB 1320 may include nonvolatile memories. The nonvolatile memory means a storage medium that stores and holds information even when power is not supplied and can reuse the stored information when power is supplied. The nonvolatile memory may include, for example, at least one of a flash memory, a hard disk, a Solid State Drive (SSD), a multimedia card micro memory, a card memory (e.g., SD or XD memory, etc.), a Read Only Memory (ROM), a magnetic memory, a magnetic disk, or an optical disk.

Referring to fig. 3, it is illustrated that a session time DB 1310 and a session usage history statistics DB 1320 are included in the memory 1300, but not limited thereto. According to an embodiment of the present disclosure, at least one of the session time DB 1310 or the session usage history statistics DB 1320 may be configured in a database form independent of the memory 1300. According to another embodiment of the present disclosure, at least one of the session time DB 1310 or the session usage history statistics DB 1320 may be connected with the electronic device 1000 through the communication interface 1100 via wired or wireless communication as an external device or a component of an external server instead of as a component of the electronic device 1000.

Fig. 4 is a flowchart of a method of operation of an electronic device according to an embodiment of the present disclosure.

Referring to fig. 4, in operation S410, the electronic device 1000 may output a response message to an input message of a user. According to embodiments of the present disclosure, the electronic device 1000 may identify domains, intents, and entities associated with an input message by analyzing the user's input message using a natural language understanding model. According to embodiments of the present disclosure, the electronic device 1000 may parse text included in an input message in units of morphemes, words, or phrases, and infer meanings of words extracted from the parsed text using language features (e.g., grammar elements) of the morphemes, words, or phrases by using a natural language understanding model. The electronic device 1000 can identify domains, intents, and entities by comparing the inferred meaning of words to each of the predetermined domains, intents, and entities provided by the natural language understanding model.

The electronic device 1000 can output a set of response messages corresponding to the domains, intents, and entities identified from the input message. According to an embodiment of the present disclosure, the electronic device 1000 may generate a response message for solving problems related to a domain, an intention, and an entity using a natural language generation model, and output the generated response message.

In operation S420, the electronic device 1000 may determine a default session time for which the session is maintained based on the difficulty level of the output response message. According to embodiments of the present disclosure, the electronic device 1000 may obtain an output tag value of the difficulty level by applying a response message as an input to the deep neural network model, and may determine a session time mapped to correspond to the obtained tag value as a default session time. The deep neural network model may be an artificial intelligence model trained by supervised learning, with a plurality of training messages previously obtained applied as input and the labeled value of the difficulty level applied as output truth value.

In operation S430, the electronic device 1000 may determine an additional session time based on the dialog history information before outputting the response message. In accordance with an embodiment of the present disclosure, in operation S410, the electronic device 1000 may determine whether to provide additional session time based on at least one of: the number of response messages previously output before outputting the response message, or session use time used in a scene performed by the previously output response message. According to an embodiment of the present disclosure, a preset ratio of a sum of session use times for each scene may be determined as the additional session time.

In operation S440, the electronic device 1000 may wait for the user to input the additional message during a session duration, which is a sum of a default session time and an additional session time.

In operation S450, the electronic device 1000 may determine the end of the session based on whether an additional message is input. According to an embodiment of the present disclosure, the electronic device 1000 may confirm whether the session duration has elapsed, and output a session end notification message based on the confirmed result.

The electronic device 1000 may output a preset session end notification message for the response message, but is not limited thereto. According to an embodiment of the present disclosure, the electronic device 1000 may generate a session end notification message asking whether a problem related to an intention and an entity identified from a response message is solved by using a natural language generation model, and output the generated response message.

Fig. 5a is a diagram illustrating a method performed by an electronic device for training a deep neural network model for classifying a difficulty level of a response message, according to an embodiment of the present disclosure.

Referring to fig. 5a, the electronic device 1000 may train by applying a plurality of training messages 510-1 through 510-n (hereinafter also referred to as a first training message 510-1 and a second training message 510-2) as inputs and applying difficulty tag values 520-1 through 520-n of each of the plurality of training messages 510-1 through 510-n as output true values. As previously obtained messages for training the deep neural network model 1214, the plurality of training messages 510-1 through 510-n may include examples of response messages of the chat bot to the user input message. The plurality of training messages 510-1 through 510-n may include a context related to requesting, asking, replying to, or providing information. For example, the plurality of training messages 510-1 through 510-n may be "Galaxy S7 LTE 128G has a price of 762,000 Korean". Or do you check if the power cord is plugged? ".

The plurality of training messages 510-1 through 510-n may be input to the embedding module 1212 before being input to the deep neural network model 1214. The embedding module 1212 may parse the input text and extract at least one term therefrom, and may quantize the extracted at least one extracted term into a vector. For example, the embedding module 1212 may convert at least one Word into an embedded vector by using a known machine learning model such as Word2vec, gloVe, or onehot encoding. However, the embedding model used by the embedding module 1212 is not limited to the above examples.

Processor 1200 (referring to fig. 3) of electronic device 1000 may extract at least one word by parsing the plurality of training messages 510-1 through 510-n in units of words using embedding module 1212 and convert the extracted word into an embedded vector. The processor 1200 may arrange the embedded vectors of at least one word in a matrix form. In the embodiment of the present disclosure shown in fig. 5a, a first training message 510-1 may be converted to a first embedded vector by the embedding module 1212, and a second training message 510-2 may be converted to a second embedded vector by the embedding module 1212.

The processor 1200 may input the embedded vector to the deep neural network model 1214. In this case, the embedded vector may be input into the deep neural network model 1214 as a feature vector.

The deep neural network model 1214 is an artificial intelligence model configured to perform training by using the embedding vectors provided from the embedding module 1212. In accordance with embodiments of the present disclosure, the deep neural network model 1214 may be trained by supervised learning, with the embedded vectors applied as inputs and the difficulty marker values 520-1 through 520-n applied as true values. Here, the "difficulty level" refers to the difficulty level of solving a problem by a request, a problem, a response, or information provided by the plurality of training messages 510-1 to 510-n. In the presently disclosed embodiment shown in fig. 5a, the difficulty level may be an integer value ranging from 0 to n, but is not limited thereto. The "difficulty level flag value" refers to a value preset with respect to difficulty. For example, the price of the first training message 510-1"Galaxy S7 LTE model is 762,000 Korean (won). "can be mapped to difficulty level 0, so a first flag value 520-1 with difficulty level 0 can be trained to output a true value, and a second training message 510-2" do you check if the power line is plugged? "may be mapped to difficulty level 1, and therefore, the second marker value 520-2 with difficulty level 1 may be trained to output a true value.

According to embodiments of the present disclosure, the deep neural network model 1214 may include multiple hidden layers, which are inner layers that perform operations. The deep neural network model 1214 may include, for example, at least one of a CNN, a recurrent neural network model (RNN), RBM, DBN, BRDNN, or a deep Q network, but is not limited to the examples described above. However, the deep neural network model 1214 is not limited to the above example, and may include all known neural network models based on deep learning.

When the deep neural network model 1214 is implemented as a CNN, by using a filter having a preset size and a preset number of channels, a feature value can be extracted from an embedded vector applied as an input, a plurality of layers including the extracted feature value can be obtained, and a feature vector map can be obtained by applying weights to the plurality of layers. In the process of obtaining the feature vector diagram, a rectified linear unit (Relu) model may be used, and in order to improve efficiency, a training model may be regularized by dropping (dropout), and an operation of performing pooling (pooling) or maximizing pooling may also be added in the process. The eigenvalues obtained by pooling or max-pooling are then merged by the full connected layer and can be trained to output a signature related to the compression ratio by an activation function comprising softmax, sigmoid and hyperbolic tangent. The values of parameters included in the deep neural network model 1214 may be changed by training the deep neural network model 1214. For example, the weights and bias values of the layers included in the deep neural network model 1214 may be changed.

Fig. 5b is a diagram illustrating a method performed by an electronic device to obtain difficulty level information about a response message by using a deep neural network model according to an embodiment of the present disclosure.

Referring to fig. 5b, the electronic device 1000 may output a response message 530 for an input message of the user. The output response message 530 may be input by the electronic device 1000 to the embedding module 1212 before being input to the deep neural network model 1214. When text included in the response message 530 is input to the embedding module 1212, the embedding module 1212 may extract at least one word by parsing the input text in units of words and may output an embedding vector by quantizing the extracted at least one word. The embedded vectors output by the embedding module 1212 may be input to the deep neural network model 1214.

As shown in FIG. 5a, the deep neural network model 1214 is an artificial intelligence model trained by supervised learning, where a plurality of training messages 510-1 through 510-n (see FIG. 5 a) are applied as inputs and difficulty marker values 520-1 through 520-n (see FIG. 5 a) are applied as output truth values. When the embedded vector converted from the response message 530 is input to the deep neural network model 1214, a flag value predicted as the difficulty level of the response message 530 may be output. In the embodiment of the present disclosure shown in fig. 5b, the price of response message 530"galaxy Z fold 5g 256g model is 2,199,000 won. "difficulty level may be predicted to be 0, and a marker value of difficulty level 0 may be output through the deep neural network model 1214.

Fig. 5c is a diagram illustrating a method performed by an electronic device to determine a default session time based on a difficulty level in accordance with an embodiment of the present disclosure.

Referring to fig. 5c, a flag value of the difficulty level of the response message 530 (refer to fig. 5 b) output through the deep neural network model 1214 (refer to fig. 5 b) may be input to the default session time determination module 1210. The default session time determination module 1210 is a module configured to obtain information about the difficulty level of the response message 530 and determine a default session time based on the difficulty level of the response message 530. According to an embodiment of the present disclosure, the default session time determination module 1210 may search the session time DB 1310 for a session time according to difficulty.

The session time DB 1310 is a database storing session times according to difficulty levels. According to an embodiment of the present disclosure, the session time DB 1310 may store data of a plurality of difficulty mark values 1312-1 to 1312-n (hereinafter also referred to as first mark value 1312-1 and n-th mark value 1312-n) and a plurality of session times 1314-1 to 1314-n (hereinafter also referred to as first session time 1314-1 and n-th session time 1314-n). The plurality of session times 1314-1 through 1314-n may be mapped to correspond to each of the plurality of difficulty marker values 1312-1 through 1312-n. According to an embodiment of the present disclosure, the session time DB 1310 may store data for a plurality of difficulty tag values 1312-1 through 1312-n and a plurality of session times 1314-1 through 1314-n as key-value types. In the embodiment of the present disclosure shown in fig. 5c, a first flag value 1312-1 with a difficulty level of 0 and a first session time 1314-1 of 30 seconds may be stored as a key-value type in the session time DB 1310. Also, an nth tag value 1312-n of difficulty level n and an nth session time 1314-n of 90 minutes may be stored as a key-value type in the session time DB 1310.

The electronic device 1000 may search for a session time from the session time DB 1310 according to the difficulty level of the response message and provide information about the session time to the electronic device 1000 by using the default session time determination module 1210. In the embodiment shown in fig. 5c, the default session time determination module 1210 may receive an input of a flag value of the difficulty level of the response message 530 (refer to fig. 5 b), and may search for a first session time 1314-1, which is a session time corresponding to the flag value of the difficulty level 0 in the session time DB 1310. The first session time 1314-1 may be 30 seconds and the default session time determination module 1210 may obtain information about the session time (30 seconds) from the session time DB 1310.

Fig. 6 is a flowchart illustrating an operation method of an electronic device according to an embodiment of the present disclosure.

Referring to fig. 6, operations S610 and S620 among the operations shown in fig. 6 are details of operation S410 shown in fig. 4. Operations S630 to S660 shown in fig. 6 are details of operation S420 shown in fig. 4.

In operation S610, the electronic device 1000 may determine a response message mapped to correspond to the input message. The electronic device 1000 may identify domains, intents, and entities of an input message by analyzing the input message received from a user. According to embodiments of the present disclosure, the electronic device 1000 may parse text in units of morphemes, words, or phrases by using a natural language understanding model, and infer the meaning of the parsed text by using linguistic features (e.g., syntax elements) of the morphemes, words, or phrases. The electronic device 1000 can identify domains, intents, and entities by comparing the inferred meaning of words to each of the predetermined domains, intents, and entities provided by the natural language understanding model.

The electronic device 1000 may output a response message regarding the identified domain, intent, and entity. According to embodiments of the present disclosure, the response message may be preset to correspond to a domain, an intention, and an entity. However, the present disclosure is not limited thereto, and the electronic device 1000 may generate response messages related to domains, intents, and entities by using a natural language generation model.

In operation S620, the electronic device 1000 may output the determined response message.

In operation S630, the electronic device 1000 may recognize whether a default session time for the response message is set.

In operation S640, when the default session time for the response message is not set, the electronic device 1000 may output a flag value of the difficulty level of the response message through the deep neural network model. Here, the "deep neural network model" is an artificial model trained by supervised learning, in which a plurality of training messages are applied as inputs, and a flag value of a difficulty level set with respect to each of the plurality of training messages is applied as an output true value. The deep neural network model is a model trained before receiving an input message from a user in operation S610. The electronic device 1000 may obtain a tag value predicted as a difficulty level of the response message by applying the response message as an input to the deep neural network model.

In operation S650, the electronic device 1000 may determine a session time corresponding to the obtained tag value of the difficulty level as a default session time. According to an embodiment of the present disclosure, the electronic device 1000 may search the session time DB 1310 (refer to fig. 5 c) via the deep neural network model by using the obtained tag value of the difficulty level to obtain the session time mapped to correspond to the tag value. The electronic device 1000 may determine the session time obtained from the session time DB 1310 as a default session time.

The default session time may be defined in terms of information, responses, or requests provided by the response message. When the default session time of the response message is set, the electronic device 1000 may determine the preset session time as the default session time in operation S660.

Fig. 7 is a time chart 700 illustrating a relationship between a default session time and a session use time according to an input message input by a user and a response message provided by a chat bot according to an embodiment of the disclosure.

Referring to fig. 7, when the user is at a first point in time t ₁ When the first input message 711 is input, the chat robot may at a second point in time t ₂ A first response message 721 for the first input message 711 is output (hereinafter, the first response message 721 and the second response message 722 may also be referred to as a previously output response message 721 and a previously output response message 722). The first default session time 731 may be provided for a predetermined time from the second time point t2 at which the first response message 721 is output. A detailed method of determining the default session time has been described with reference to fig. 5c and 6, and thus redundant description is omitted. The user may at a third point in time t before the first default session time 731 elapses ₃ The second input message 712 is entered and in this case the session time may be updated.

The first session usage time 741 refers to a second time point t at which the first response message 721 is provided to the user ₂ Thereafter, the time taken by the user to perform a particular operation based on the request, reply, or information provided by the first response message 721, or until a third point in time t at which the second input message 712 is entered for additional queries or additional requests ₃ The time used. For example, the first default session time 731 may be 5 minutes and the first session usage time 741 may be 4 minutes. When the session use time exceeds the default session timeThe session may terminate.

The chat robot may at a fourth point in time t ₄ A second response message 722 associated with the second input message 712 is output. In the same manner as the first default session time 731, the second default session time 732 may be provided for outputting the fourth point in time t of the second response message 722 from ₄ A preset time from the start. The user may be at a fifth point in time t before the second default session time 732 elapses ₅ The third input message 713 is input and in this case the session time may be updated.

The second session use time 742 refers to the time taken for the user to perform a specific operation according to the request, reply, or information provided by the second response message 722 after the fourth time point t4 of the second response message 722 provided to the user, or until the fifth time point t of the third input message 713 for additional inquiry or additional request ₅ The time used. For example, the second default session time 732 may be 10 minutes and the second session use time 742 may be 6 minutes.

"scenario" refers to a problem-solving operation (or process) provided by a chat bot for a user's input message, such as requesting, asking, answering, or providing information. According to embodiments of the present disclosure, a scenario may be defined as a point in time when a response message is provided to a user to a point in time when an additional input message of the user is input. In the embodiment of the present disclosure shown in fig. 7, the first scenario may refer to a second point in time t at which the first response message 721 is output ₂ And a third point in time t at which the second input message 712 is input ₃ Time interval between, and the second scenario may refer to a fourth point in time t at which the second response message 722 is output ₄ And a fifth time point t at which the third input message 713 is input ₅ Time interval between.

The electronic device 1000 may determine the additional session time based on information related to at least one of: the number of response messages previously output before the current output response message is output, or the session use time used in a scene performed by the previously output response message. According to embodiments of the present disclosure, when the number of previously output response messages is one or more, the electronic device 1000 may determine whether to provide additional session time. A preset ratio of the sum of session use times for each scene may be determined as the additional session time. Detailed embodiments of the present disclosure in which the electronic device 1000 determines additional session times will be described with reference to fig. 8-10.

Fig. 8 is a flowchart illustrating a method performed by an electronic device to determine whether to provide additional session time, according to an embodiment of the present disclosure.

Referring to fig. 8, operations S810 to S830 shown in fig. 8 are details of operation S430 shown in fig. 4. Operation S810 may be performed after performing operation S420 shown in fig. 4. Operation S440 shown in fig. 4 may be performed after performing operations S820 and S830.

In operation S810, the electronic device 1000 may recognize whether there is a previously output response message before the response message.

In operation S820, when it is confirmed that there is a previously output response message, the electronic device 1000 may determine a certain ratio of the sum of session use times for each scene as an additional session time according to the previously output response message. Referring to fig. 7, when the currently output response message is a third response message 723 (refer to fig. 7) (hereinafter, also referred to as a current response message 723 or a currently output response message 723), since two response messages including a first response message 721 (refer to fig. 7) and a second response message 722 (refer to fig. 7) are output, the electronic device 1000 can determine whether to provide additional session time. Further, since the first session use time 741 (refer to fig. 7) and the second session use time 742 (refer to fig. 7) are used for the first scene and the second scene, respectively, the electronic device 1000 may determine a certain ratio (e.g., 20%) of the sum of the first session use time 741 and the second session use time 742 as the additional session time. For example, when the first session use time 741 is 4 minutes and the second session use time 742 is 6 minutes, the electronic device 1000 may determine 2 minutes (i.e., 20% of the sum of the first session use time 741 and the second session use time 742 of 10 minutes) as the additional session time of the third response message 723 (refer to fig. 7).

In operation S830, when it is confirmed that there is no previously output response message (i.e., when the currently output response message is the first response message), the electronic device 1000 may determine not to provide the additional session time.

Fig. 9 is a flowchart illustrating a method performed by an electronic device to determine whether to provide additional session time, according to an embodiment of the present disclosure.

Referring to fig. 9, operations S910 to S940 shown in fig. 9 are details of operation S430 shown in fig. 4. Operation S910 may be performed after performing operation S420 shown in fig. 4. Operation S440 shown in fig. 4 may be performed after performing operations S930 and S940.

In operation S910, the electronic apparatus 1000 may calculate a session time usage ratio for each scene. According to embodiments of the present disclosure, the electronic device 1000 may calculate the session time usage ratio by performing an arithmetic operation of dividing the session usage time by the default session time. For example, when the default session time is 10 minutes and the session use time is 6 minutes, the session time use ratio is 60%.

According to embodiments of the present disclosure, the electronic device 1000 may calculate a session time usage ratio for each scene according to a previously output response message. Referring to fig. 7, when the currently output response message is a third response message 723 (refer to fig. 7), the electronic device 1000 may calculate a first session time usage ratio for a first scene and a second session time usage ratio for a second scene. For example, when the first default session time 731 (refer to fig. 7) is 5 minutes and the first session use time 741 (refer to fig. 7) is 4 minutes, the first session use ratio is 80%; and when the second default session time 732 (refer to fig. 7) is 10 minutes and the second session use time 742 (refer to fig. 7) is 6 minutes, the second session time use ratio is 60%.

In operation S920, the electronic device 1000 may compare the session time usage ratio with a preset threshold α. The electronic device 1000 may determine whether the session time usage ratio exceeds a threshold α.

In operation S930, the electronic device 1000 may determine to provide the additional session time when the session time usage ratio exceeds the preset threshold α. For example, when the preset threshold α is 50%, since the first session time usage ratio is 80% and the second session time usage ratio is 60%, the electronic device 1000 may determine that the session time usage ratio has exceeded the preset threshold α and determine to provide additional session time.

The additional session time may be determined as a certain ratio of the sum of session usage times for each scene. For example, when the first session use time is 4 minutes and the second session use time is 6 minutes, it may be determined that the additional session time is 2 minutes, that is, 20% of the sum of the first session use time and the second session use time.

In operation S940, the electronic device 1000 may determine that the additional session time is not provided when the session time usage ratio is equal to or less than the preset threshold α.

Fig. 10 is a flowchart illustrating a method performed by an electronic device to determine whether to provide additional session time, according to an embodiment of the present disclosure.

Referring to fig. 10, operations S1010 to S1040 shown in fig. 10 are details of operation S430 shown in fig. 4. Operation S1010 may be performed after performing operation S420 shown in fig. 4. Operation S440 shown in fig. 4 may be performed after performing operations S1030 and S1040.

In operation S1010, the electronic device 1000 may calculate an average value of session use times before outputting the response message. Referring to fig. 7, when the currently output response message is a third response message 723 (refer to fig. 7), the electronic device 1000 may calculate an average value of session use times in the first scene and an average value of session use times in the second scene. For example, when the first session use time 741 (refer to fig. 7) used in the first scene is 4 minutes and the second session use time 742 (refer to fig. 7) used in the second scene is 6 minutes, an average value of 5 minutes may be calculated.

In operation S1020, the electronic device 1000 may compare the calculated average value with a preset threshold t _th Comparison is performed to determine an average valueWhether or not the threshold t is exceeded _th 。

In operation S1030, when the average value exceeds the preset threshold t _th When the electronic device 1000 may determine to provide additional session time. For example, when a threshold t is preset _th At 3 minutes, since the average value calculated in operation S1010 is 5 minutes, the electronic device 1000 may determine to provide the additional session time.

In operation S1040, when the average value is equal to or smaller than the preset threshold t _th When the electronic device 1000 may determine not to provide additional session time.

The more previously output response messages (refer to fig. 7) that exist before the current output response message 723, the more burden the user has to end the session. When it takes time due to actions taken by the user according to the current response message 723 (e.g., checking for a fault, identifying a device status, etc.), or when the user has not entered an additional message due to a mistake, the default session time may elapse, and the session may end. In this case, the user may need to repeatedly input the same input message to realize the same response message 723, which is cumbersome and reduces the user's convenience.

The electronic device 1000 according to the embodiment shown in fig. 8 to 10 may provide additional session time based on the number of previously output response messages 721 and 722 (refer to fig. 7) before the currently output response message 723 and information according to at least one of session use time of each scene of the previously output response messages 721 and 722, thereby solving the problem of unexpected end of a session due to the default session time passing. In addition, by considering the session use time used by the user to implement the current response message 723, an optimized additional session time can be provided for each user, thereby improving user convenience.

Fig. 11 is a flowchart illustrating a method of outputting a session end notification performed by an electronic device according to an embodiment of the present disclosure.

Referring to fig. 11, operations S1110 to S1150 shown in fig. 11 are details of operation S450 shown in fig. 4. Operation S1110 may be performed after performing operation S440 shown in fig. 4.

In operation S1110, the electronic apparatus 1000 may recognize whether the session duration has elapsed. According to embodiments of the present disclosure, the session duration may be determined by the sum of the default session time and the additional session time. The electronic device 1000 may identify whether additional messages of the user are input during the duration of the session from when the response message is output.

In operation S1120, the electronic device 1000 may output the session end notification 250. According to embodiments of the present disclosure, the electronic device 1000 may generate a message informing a user that the session duration has ended using a natural language generation model by using an intention and an entity related to the response message, and may output the generated response message. The session end notification message may be, for example, "is the problem with the refrigerator solved? The chat is ended. "

In operation S1130, the electronic apparatus 1000 may recognize whether a response message of the user to the end notification is input. The reply message of the user may be a message indicating whether the problem has been solved with Y/N (such as "yes" or "no"), but is not limited thereto. The reply message may be a message related to an additional question or an additional request.

In operation S1140, the electronic device 1000 may update the session duration when confirming that the response message of the user is input.

In operation S1150, when it is confirmed that the answer message of the user is not input, the electronic device 1000 may determine to end the session. According to an embodiment of the present disclosure, even if a response message is input, when the response message is a message notifying that a problem has been solved (for example, is the chat ended for "is a problem about refrigerator solved.

In the embodiment of the present disclosure shown in fig. 11, the electronic device 1000 may not immediately end the session when the session duration has elapsed, and may output a session end notification message, thereby notifying the user whether the session will end, and allowing the user to input additional messages if necessary. Accordingly, the electronic device 1000 of the present disclosure can prevent the problem of accidental end of a session and improve user convenience.

Fig. 12 is a flowchart illustrating a method performed by an electronic device to adjust a default session time, according to an embodiment of the present disclosure.

Referring to fig. 12, in operation S1210, the electronic device 1000 may store statistical data regarding session use history information of a plurality of users. According to embodiments of the present disclosure, the session use history information may be statistics of a plurality of users regarding at least one of: the number of session extension requests, the number of scenarios performed before outputting the response message, or the session usage time. According to embodiments of the present disclosure, the electronic device 1000 may classify and store session usage history information of each response message.

In operation S1220, the electronic device 1000 may adjust the default session time based on the session use history information. The electronic device 1000 may extend or shorten the default session time based on session usage history information of a plurality of users related to the currently output response message. A detailed embodiment of the present disclosure in which the electronic device 1000 extends or shortens the default session time will be described with reference to fig. 13.

Referring to fig. 13, in operation S1310, the electronic device 1000 may recognize whether there is a request to extend a session in a previous session. According to an embodiment of the present disclosure, the electronic device 1000 may search for statistical data on session use history information related to a currently output response message from session use history information of a plurality of users stored in the session use history statistical data DB 1320 (refer to fig. 3). The session use history statistical data DB 1320 is a database for storing statistical data on session use histories of a plurality of users. Based on the search results of the session use history statistics DB 1320, the electronic device 1000 can obtain information on whether or not a plurality of users have requested session extension in the previous session before outputting the response message.

In operation S1320, when confirming that there is a session extension request, the electronic device 1000 may increase the session extension number by 1.

In operation S1330, the electronic device 1000 may determine whether the session extension number exceeds α% of the total number of sessions.

In operation S1340, when it is confirmed that the number of session extension times exceeds α% of the total number of sessions, the electronic device 1000 may extend the default session time of the response message.

In operation S1350, the electronic device 1000 may update the session time when confirming that the session extension number is α% or less of the total number of sessions.

In operation S1360, when it is confirmed that there is no session extension request in the previous session, the electronic device 1000 may determine whether the maximum value of the session use time for the response message is less than β% of the default session time. According to an embodiment of the present disclosure, the electronic device 1000 may obtain information on session use times of a plurality of users of a currently output response message from the session use history statistics DB 1320. The electronic apparatus 1000 may identify the maximum value of the session use time by the information on the session use times of the plurality of users obtained from the session use history statistical data DB 1320 and determine whether the maximum value is less than β (e.g., 50%) of the default session time.

In operation S1370, when the maximum value of the session use time is less than β%, the electronic device 1000 may shorten the default session time of the response message. The electronic device 1000 may shorten the default session time by, for example, 25%.

The electronic device 1000 according to the embodiment of the present disclosure shown in fig. 12 and 13 may store session use history information about other users of the response message and lengthen or shorten a default session time by using the stored session use history information of the plurality of users, thereby having a technical effect of adaptively adjusting the default session time by reflecting the plurality of user experiences.

The programs executed by the electronic device 1000 described herein may be implemented as hardware components, software components, and/or a combination of hardware and software components. The program may be executed by any system capable of executing computer-readable instructions.

The software may include computer programs, code, instructions, or combinations thereof, and may configure the processing devices to operate as desired, or to command the processing devices independently or collectively.

The software may be implemented as a computer program comprising instructions stored in a computer readable storage medium. Examples of the computer readable recording medium may include magnetic storage media (e.g., ROM, RAM, floppy disks, and hard disks) and optical recording media (e.g., compact disk-ROM (CD-ROM) and Digital Versatile Disks (DVDs)). The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. The medium may be readable by a computer, stored in a memory, and executed by a processor.

The computer readable storage medium may be provided in the form of a non-transitory storage medium. The term "non-transitory" simply means that the storage medium is a tangible device and does not include a signal, but the term does not distinguish between a case where data is semi-permanently stored in the storage medium and a case where data is temporarily stored in the storage medium.

The programs according to the embodiments of the present disclosure disclosed in the present specification may be included in and provided in a computer program product. The computer program product may be traded as a product between the buyer and the seller.

The computer program product may include a software (S/W) program and a non-transitory computer-readable recording medium storing the S/W program. For example, the computer program product may include a program product that is executed by a device or electronic marketplace (e.g., google Play ^TM Mall, app businessCity, etc.) and a software program type product (e.g., downloadable application) that is electronically distributed. For electronic distribution, at least a portion of the software program may be stored in a storage medium or temporarily generated. In this case, the storage medium may be a storage medium of a manufacturer server, a server of an electronic market, or a relay server that temporarily stores a software program.

The computer program product may comprise a storage medium of a server or a storage medium of a device in a system comprising the server and the device. Alternatively, when there is a third device (e.g., a smart phone) in communication with the server or device, the computer program product may include a storage medium for the third device. Alternatively, the computer program product may comprise a software program transferred from the server to the device or the third device or from the third device to the device.

In this case, one of the server, the device, and the third device may perform the method according to the embodiments of the present disclosure by executing the computer program product. Alternatively, two or more of the server, the device and the third device may execute the computer program product in order to perform the method according to embodiments of the present disclosure in a distributed manner.

For example, a server may execute a computer program product stored in the server to control a device communicatively connected to the server to perform a method according to an embodiment of the present disclosure.

In another example, a third device may execute a computer program product to control an apparatus in communication with the third device to perform a method according to an embodiment of the present disclosure.

When the third device executes the computer program product, the third device may download the computer program product from the server and execute the downloaded computer program product. Alternatively, the third device may execute the computer program product provided in a free-loaded state and perform the method according to embodiments of the present disclosure.

While the present disclosure has been shown and described with reference to various embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present disclosure as defined by the appended claims and their equivalents.

Claims

1. A method of determining a conversation duration of a chat robot, the method comprising:

outputting a response message for the input message of the user;

determining a default session time for which the session is maintained based on the difficulty level of the outputted response message;

determining an additional session time based on session history information by the chat robot before outputting the response message;

waiting for a user to enter an additional message during a session duration, the session duration being a sum of the default session time and the additional session time; and

An end of the session is determined based on whether the additional message is entered.

2. The method of claim 1, wherein determining the default session time comprises:

obtaining a marking value of the difficulty level by applying the response message as an input to a deep neural network model; and

a session time mapped to correspond to the obtained flag value is determined as the default session time.

3. The method according to claim 2,

wherein the deep neural network model is trained by supervised learning,

wherein a plurality of training messages are applied as inputs, an

Wherein the marked value of the difficulty level is applied as an output true value.

4. The method of claim 1, wherein the conversation history information includes information about at least one of: the number of previously output response messages in a session before outputting the response message, or a session use time in a scene performed according to the previously output response message.

5. The method of claim 4, wherein the additional session time is determined by a preset ratio of the sum of session usage times for each scene.

6. The method of claim 1, further comprising:

storing session use history information of a plurality of users with respect to the outputted response message; and

the default session time is adjusted based on stored session usage history information for a plurality of users.

7. The method of claim 6, wherein the session usage history information includes statistics of the plurality of users regarding at least one of: the number of session extension requests at a point of time before outputting the response message, the number of scenes performed before the response message, or the session use time for the response message.

8. An electronic device for determining a duration of a conversation of a chat robot, the electronic device comprising:

a communication interface configured to perform transmission and reception of data with other devices;

a memory; and

a processor configured to execute one or more instructions,

wherein the processor is further configured to execute the one or more instructions to:

an input message entered by a user is obtained through the communication interface,

outputting a response message to the user's input message,

Based on the difficulty level of the outputted response message, a default session time for which the session is maintained is determined,

based on conversation history information through the chat robot before outputting the response message, determining an additional conversation time,

waiting for a user to enter an additional message during a session duration that is the sum of the default session time and the additional session time, an

9. The electronic device of claim 8, wherein the processor is further configured to execute the one or more instructions to:

obtaining an output marking value of the difficulty level by applying the response message as input to a deep neural network model; and

10. An electronic device according to claim 9,

wherein the deep neural network model is trained by supervised learning,

wherein a plurality of training messages are applied as inputs, an

11. The electronic device of claim 8, wherein the conversation history information includes information about at least one of: the number of previously output response messages in a session before outputting the response message, or a session use time in a scene performed according to the previously output response message.

12. The electronic device of claim 11, wherein the processor is further configured to execute the one or more instructions to: the additional session time is determined by a preset ratio of the sum of the session use times for each scene.

13. The electronic device of claim 8, wherein the processor is further configured to execute the one or more instructions to:

storing session use history information of a plurality of users regarding the outputted response message in a database of the memory, and

14. The electronic device of claim 13, wherein the session usage history information comprises statistics of the plurality of users regarding at least one of: the number of session extension requests at a point of time before outputting the response message, the number of scenes performed before the response message, or the session use time for the response message.

15. At least one non-transitory computer program product comprising a computer-readable storage medium, wherein the computer-readable storage medium comprises instructions for execution by a device to:

outputting a response message for the input message of the user;