KR102618335B1 - Device, method, system and computer readable storage medium to generate learning data for machine learning-based chatbot - Google Patents

Abstract

A method of generating learning data for a machine learning-based chatbot is explained. An exemplary learning data generation method includes receiving, from a user, input data including a message set including at least one message and a system response value for a dialog according to the message set; For each of the at least one message, determining one or more label data used for training the chatbot, based on a system function related to the type of dialog; and generating learning data based on the determined one or more label data.

Description

Device, method, system and computer-readable storage medium for generating learning data for machine learning-based chatbot {DEVICE, METHOD, SYSTEM AND COMPUTER READABLE STORAGE MEDIUM TO GENERATE LEARNING DATA FOR MACHINE LEARNING-BASED CHATBOT}

The present disclosure relates to an apparatus, method, and computer-readable storage medium for generating training data for a machine learning-based chatbot.

Unless otherwise stated herein, the material described in this section is not prior art to the claims in this application and should not be acknowledged as prior art by virtue of being described in this section.

As artificial intelligence techniques have recently developed, their applications are also broadening. In particular, chatbots, which enable conversations with users using natural language, are being used in various fields, and the technology is gradually developing. Conventional chatbots, such as IBM's Watson chatbot platform, use the rule-based dialog method. The rule-based dialog method can set various expected conversation scenarios for the chatbot and allow the chatbot to conduct a conversation with the user according to those expected conversation scenarios. Chatbots that use this rule-based dialog method are relatively small in size and demonstrate stable performance for preset types of conversations. However, chatbot builders must set up complex rules to encompass various types of conversations, and when adding new types of conversations, they must ensure that there are no conflicts with existing rules, which can be difficult. Additionally, chatbots conduct conversations according to set rules, so the flexibility of conversations is low.

To improve this rule-based dialog method, machine learning-based chatbots are being researched. Machine learning-based chatbots use machine learning, such as deep learning methods, in various processes such as the intent of the message in the dialog, the current situation in the dialog of the message, and the chatbot's policy decision in response to such message. While existing rule-based dialogue chatbots partially use machine learning techniques to understand the user's natural language messages and proceed with the dialogue according to set rules, machine learning-based chatbots use machine learning even in the dialogue process. This increases the flexibility of the dialog, and since additional learning needs to be performed in response to new dialog types, configuration becomes relatively simple. However, a large amount of learning data is required for machine learning-based chatbots to operate with sufficient performance. This learning data includes labels for each data, and the chatbot builder needs to directly set the labels of the learning data to ensure that the output of the machine learning model is corrected for its input.

However, learning data to correct all individual movements of a machine learning-based chatbot usually requires many types of data and labels, and the chatbot builder requires a large amount of learning data and many types of labels and data for each learning data. Writing it yourself can be a very time-consuming and inefficient task. Additionally, if multiple chatbot builders create labels, labeling may become inconsistent, which may reduce the effectiveness of chatbot learning.

The present disclosure is intended to solve the above problems and presents an apparatus, method, and computer-readable storage medium for generating learning data for a machine learning-based chatbot.

In some embodiments of the present disclosure, a learning data generation method for generating training data for a machine learning-based chatbot is described. An exemplary learning data generation method includes receiving, from a user, input data including a message set including at least one message and a system response value for a dialog according to the message set; For each of the at least one message, determining one or more label data used for training the chatbot, based on a system function related to the type of dialog; and generating learning data based on the determined one or more label data.

In some examples, system functions may be included in the input data. In some other examples, the method of generating training data further includes determining a system function based on at least one of a message set and a system response value, after receiving input data and before generating one or more label data. can do. In some additional examples, the method of generating training data may further include receiving corrected training data from a user that corrects the training data.

In some other embodiments, an apparatus for generating learning data for a machine learning-based chatbot is described. An exemplary learning data generation device includes an input module; labeling processing module; and an output module. The input module may receive input data including a message set including at least one message from the user and a system response value for a dialog according to the message set. The labeling processing module may determine one or more label data used for chatbot learning based on a system function related to the type of dialog for each of at least one message. The output module may generate learning data based on one or more determined label data. In some examples, the learning data generating device may further include a feedback processing module that receives corrected learning data that corrects the learning data from the user.

In some further embodiments, a computer-readable storage medium storing a labeling program is described. An example labeling program can perform the generation of training data for a machine learning-based chatbot. The labeling program, when executed by a computing device, includes the following operations: receiving input data from a user, including a message set including at least one message, a system response value for a dialog according to the message set; An operation of determining, for each of at least one message, one or more label data used for learning a chatbot, based on a system function related to the type of dialog; And it may include one or more instructions to perform an operation of generating learning data based on the one or more determined label data.

The above brief summary and description of effects are merely illustrative and are not intended to limit the technical details intended in the present disclosure. By referring to the following detailed description and the accompanying drawings, additional embodiments and technical features, in addition to the above-described exemplary embodiments and technical features, may be understood.

The above-described features and other features of the present disclosure will become fully apparent from the following description with reference to the accompanying drawings. Understanding that these drawings illustrate only a few embodiments according to the present disclosure and therefore should not be considered limiting its scope, the present disclosure is described in greater detail and detail through the use of the accompanying drawings. It will be.
1A and 1B are block diagrams of a machine learning-based chatbot device related to at least some embodiments of the present disclosure.
Figure 2 is a block diagram of a learning data generating device according to at least some embodiments of the present disclosure.
3 is a block diagram of a chatbot system according to at least some embodiments of the present disclosure.
4 illustrates an example process for performing training data generation, according to at least some embodiments of the present disclosure.
5 illustrates a computer program product that can be used to perform generation of training data, in accordance with at least some embodiments of the present disclosure.
6 is an example block diagram of a computing device arranged in accordance with at least some embodiments of the present disclosure.
7 is an example of input data to a learning data generating device, according to at least some embodiments of the present disclosure.
Figure 8 is an example of learning data generated based on the input data of Figure 7.

Hereinafter, with reference to the attached drawings, implementation examples and embodiments of the present disclosure will be described in detail so that those skilled in the art can easily implement the present disclosure. However, the present application may be implemented in many different forms and is not limited to the implementation examples and examples described herein.

The present disclosure generally relates to a method, device, system, computer-readable storage medium, and programs stored therefor for generating training data for a machine learning-based chatbot.

1A and 1B are block diagrams of a machine learning-based chatbot device 100 related to at least some embodiments of the present disclosure. FIG. 1A shows an example in which the chatbot device 100 performs a dialogue with a user, and FIG. 1B shows an example in which the manager 160 of the chatbot device 100 trains the chatbot device 100. As shown in FIGS. 1A and 1B, the chatbot device 100 includes a Natural Language Understanding module (NLU module) 110 and a Dialogue State Tracking module (DST module). It may include (120), a policy module (130), a natural language generation module (hereinafter referred to as NLG module) (140), and a learning management unit (150). It will be understood by those skilled in the art that the configuration of the chatbot device 100 shown in FIGS. 1A and 1B is merely exemplary, and that various types of chatbot device 100 may be associated with the present disclosure. For example, the chatbot device 100 may have components connected in a different way than shown in FIGS. 1A and 1B, arbitrary components may be added or omitted, or it may be divided into smaller components.

The user can use the chatbot device 100 directly or use a user terminal device connected to the chatbot device 100 through a wired or wireless network to perform the chatbot and chatbot dialog implemented by the chatbot device 100. . The chatbot device 100 may receive an input message entered by a user, and this input message may be a natural language. The chatbot device 100 may output a response message in natural language in response to the input message. The chatbot device 100 may perform a dialogue with the user as follows, for example.

As shown in FIG. 1A, the chatbot device 100 can perform a dialogue with the user using the NLU module 110, DST module 120, policy module 130, and NLG module 140. At least some of the NLU module 110, DST module 120, policy module 130, and NLG module 140 may be implemented using machine learning techniques. The NLU module 110 can receive an input message from a user, and can determine the intent value and entity value of the input message based on the input message. For example, if the input message is "I would like to book a 3-star hotel on December 25th", the intent value could be "hotel reservation", while the entity value would be "hotel-star" and the entity variables "date" and It may contain the corresponding values "3 stars" and "December 25th", respectively.

The DST module 120 can search and manage dialogs performed by the chatbot device 100 and data related thereto. In some examples, DST module 120 may receive intents and entities determined by NLU module 110. The DST module 120 may acquire data about the history or context of the currently executing dialog and output a DST value based on data, intents, and entities about this context. For example, if the input message is "I would like to reserve a 3-star hotel for December 25th", the DST module 120 will perform a dialog regarding the end date of the current hotel stay, for example, taking into account the previous dialog within the dialog being performed and It can be determined that there is, and accordingly, the DST variables "hotel-star" and "to-date" and the corresponding values "3 stars" and "December 25" can be determined. These DST values may have a predetermined data format.

The policy module 130 may determine policies for actions to be performed by the chatbot device 100 and response messages after receiving the input message by the NLU module 110. These policies may include the type of response message to the input message. The policy module 130 may receive an intent value and an entity value from the NLU module 110, and may receive a DST value from the DST module 120. In some examples, the policy module 130 may call an external API (Application Program Interface) based on at least one of an intent value, an entity value, and a DST value, and determine the type of response message to be output based on this API. You can decide. In one example, the type of message determined by policy module 130 may take the form of asking the user for additional information. In another example, the type of message determined by policy module 130 may take the form of containing response data to a query to the user.

Meanwhile, the chatbot device 100 can perform various operations, such as reservation, search, and calculation, based on the policy determined by the policy module 130. The NLG module 140 may generate a response message based on the policy and response data. For example, if the policy module 130 determines the policy of asking for the start date of the trip, the NLG module 140 may generate natural language asking for the start date of the trip.

As described above, at least some of the NLU module 110, DST module 120, policy module 130, and NLG module 140 use machine learning techniques, so as shown in FIG. 1B, the chatbot device ( The manager 160 of 100 can improve the performance of the chatbot device 100 by training one or more machine learning models in the chatbot device 100 using learning data. In some examples, when the manager 160 inputs a plurality of learning data, the learning management unit 150 may receive a plurality of learning data. The learning management unit 150 may train the chatbot device 100 to learn one or more machine learning models using a plurality of received learning data. This learning may include adjusting weights or parameters within the machine learning model.

The learning data used to learn the machine learning model is a data set in a predetermined format that includes one or more data such as input data and output data, and the manager 160 of the chatbot device 100 creates a data set with a predetermined format. are required to do so. Learning data includes data input to the chatbot device 100 or output internally or externally to the chatbot device 100, such as an input message, an intent value, one or more entity values, a DST value, and a system response value. It may include a system function related to the type of dialog being performed by the device 100. System functions can contain data associated with the dialog. For example, if the dialog is about “flight reservation,” the system function may include data regarding the flight number of the booked aircraft.

In order to train one or more machine learning models included in the chatbot device 100, the manager 160 must directly create and input learning data according to a predetermined format. For example, the manager 160 needs to include in the learning data each label corresponding to what the individual data of the above-described entity variable, DST variable, and input message represents. However, sufficient training of a machine learning model requires a large amount of training data, and it can be a very time-consuming task for administrators to create such training data directly.

Figure 2 is a block diagram of a learning data generating device 200 according to at least some embodiments of the present disclosure. Learning data generating device 200 may be a single device, such as, for example, a desktop computer, laptop computer, smartphone, tablet computer, mobile phone, personal digital assistant (PDA), special purpose device, or fusion device that includes any of the above functions. computing device; Alternatively, it may be a multiple computing configuration including multiple computing devices, such as a server farm, distributed network, or cloud computing configuration, but is not limited thereto.

In some embodiments, the learning data generating device 200 may include an input module 210, a labeling processing module 220, and an output module 230, as shown in FIG. 2 . Additionally, the learning data generating device 200 may further include a feedback processing module 240. The learning data generating device 200 may store one or more programs according to the present disclosure in an internal memory, and these one or more programs may be executed by a processor, such as an input module 210 and a labeling processing module 220, which will be described below. , at least some of the operations of the output module 230 and the feedback processing module 240 can be performed. These components 210, 220, and 230 and optional components 240 are shown as individual components, but may be separated into additional components, combined into fewer components, or eliminated, without departing from the scope of the disclosed subject matter. In addition, in Figure 2, the component is described as being implemented by software, but those skilled in the art will understand that each function and/or operation of the component can be implemented individually and/or comprehensively through hardware, software, firmware, or any combination thereof. It will be understood that the learning data generating device 200 can generate learning data for a machine learning-based chatbot, as described below.

The input module 210 may receive input data from the user. A user may need learning data to train a machine learning-based chatbot, for example, like the manager 160 in FIG. 1B. In some examples, the input data may include a message set including at least one message and a system response value for a dialog according to the message set. At least one message included in the message set may be a text message using natural language that is or can be used in a dialog performed by a chatbot. The system response value to the dialog may be the chatbot's response value to any of at least one message or the chatbot's response value to the dialog according to the message set. For example, if the dialog relates to aircraft reservations, the system response values may be the flight number and departure time.

The labeling processing module 220 may use the input data received by the input module 210 to determine one or more label data used for learning the chatbot. Each piece of label data may include data and at least one label value corresponding to the data. The labeling processing module 220 may determine one or more label data for each of at least one message included in the message set based on a system function. In some examples, a system function may be related to the type of dialog associated with the message set and may include one or more variables. In some other examples, system functions may be associated with the behavior of a chatbot.

The labeling processing module 220 may determine one or more label data for each of at least one message, based on a system function. In some examples, labeling processing module 220 may derive one or more data from each message, determine a label value for each of the one or more derived data based on a system function, and associate the derived data with the label value. You can do it. In some other examples, the labeling processing module 220 may determine a label value required for a type of dialog based on a system function and obtain data corresponding to the determined label value from the message set and dialog-related data. . One or more label data may include, for example, at least one of an intent for a message, one or more entities, and a dialog state value within a dialog. The one or more label data is not limited to this and may include any data required as a variable in the learning data.

In some embodiments, input data may include system functions. If the system function is included in the input data received by the input module 210, the labeling processing module 220 may determine the system function from the input data. As such, when a system function is included in the input data, the user needs to directly select such a system function, so this may be referred to as a “semi-automatic mode.”

Meanwhile, in additional or alternative embodiments, labeling processing module 220 may operate in an “automatic mode.” If the labeling processing module 220 operates in automatic mode, the input message may not include system functions. In this case, in some examples, labeling processing module 220 may determine a system function based on at least one of a message set of input data and a system response value. The labeling processing module 220 may then determine one or more label data based on the system function. In some additional examples, labeling processing module 220 may determine the system function further based on predetermined training data. For example, if the labeling processing module 220 determines that any one message or system response value among the predetermined learning data is the same or similar to a message or system response value included in the current input data, the labeling processing module 220 operates the system of the learning data. The function can be determined as a system function for the corresponding input data.

Whether the labeling processing module 220 operates in “semi-automatic mode” or “automatic mode” may be determined depending on implementation. In some examples, the operation mode of the labeling processing module 220 may be predetermined by the user. In some other examples, the operation mode of the labeling processing module 220 may be implemented to be automatically determined based on arbitrary conditions such as the type of the input message (i.e., whether a system function is included), the amount of prepared training data, etc. do.

The output module 230 may generate learning data based on one or more label data determined by the labeling processing module 220. The learning data generated by the output module 230 may have a predetermined format to be used as learning data used for learning of the chatbot. In some examples, when there are multiple label values corresponding to the data determined by the labeling processing module 220, the output module 230 may generate training data to include candidate items for label values corresponding to the data. there is.

In a further embodiment, the feedback processing module 240 may receive corrected training data from a user that corrects the training data generated by the output module 230. In some examples, the output module 230 may generate training data according to a corresponding format using one or more label data determined by the labeling processing module 220. Some of the label data determined by the labeling processing module 220 may include only either a label value or data, and may not include the corresponding data or label value. In this example, the output module 230 may leave data not included in the label data or label values as blank in the training data. If the learning data output by the output module 230 has blanks or is insufficient for use in a chatbot, the user can complete the labeling task for the learning data by adding the missing parts. If there is an error in the data included in the output training data, the user can correct it. The feedback processing module 240 may receive the learning data added or modified in this way as corrected learning data.

In some examples, the feedback processing module 240 determines the labeling value of the labeling processing module 220 based on the training data generated by the output module 230 and the corrected training data received by the feedback processing module 240. method can be evaluated. Thereafter, the user may modify the labeling processing module 220, system function, or learning data based on the results of the evaluation. In some other examples, feedback processing module 240 may adjust how labeling processing module 220 determines one or more labeling values based on the training data and corrected training data. In this example, the feedback processing module 240 may use an artificial intelligence module such as a machine learning module, depending on the implementation.

In this way, the learning data generating device 200 automatically generates at least part of the learning data, thereby reducing the time and cost required for the user to create and prepare machine learning-based learning data. Additionally, as the learning data generating apparatus 200 operates in an automatic mode and can form label data more accurately through feedback, the time and cost required to generate learning data can be further reduced.

Figure 3 is a block diagram of a chatbot system 300 according to at least some embodiments of the present disclosure. As shown in FIG. 3, the chatbot system 300 may include a learning data generating device 310 and a chatbot device 320. The learning data generating device 310 may generate learning data like the learning data generating device 200 described in FIG. 2 . The chatbot device 320 can implement a machine learning-based chatbot like the chatbot device 100 in FIGS. 1A and 1B, and various types of well-known chatbot algorithms can be used. In some examples, learning data generated by the learning data generating device 310 may be stored in a memory within the chatbot system 300, and the chatbot device 320 may perform learning using the learning data stored in the memory. .

In some examples, learning data generated by the learning data generating device 310 may be stored after being modified or corrected by a user. In some other examples, the learning data generated by the learning data generating device 310 may not require modification or correction by the user. In this example, the user inputs simpler input data than the learning data used in the chatbot device 320 into the learning data generating device 310, thereby enabling the chatbot device 320 to learn. The learning data generating device 310 may be integrally formed as the chatbot device 320 and the chatbot system 300, but may be implemented as a separate device from the chatbot device 320 to generate data generated by the learning data generating device 310. It is also possible to implement learning data to be manually input into the chatbot device 320 by the user.

4 illustrates an example process 400 for performing training data generation, in accordance with at least some embodiments of the present disclosure. For example, process 400 may be performed under the control of a computing device, such as learning data generating device 200 and/or chatbot system 300 of FIG. 2 . Accordingly, the process 400 may generate training data for a machine learning-based chatbot, such as the chatbot device 100 of FIGS. 1A and 1B, as described below. Process 400 shown in FIG. 4 may include one or more operations, functions or actions as illustrated by blocks 410, 420, 430 and/or 440. The schematic operations illustrated in FIG. 4 are provided by way of example only, and some of the operations may be optional, combined into fewer operations, or expanded into additional operations without departing from the essence of the disclosed embodiments. there is. Process 400 may begin at block 410, which receives input data.

At block 410, the computing device may receive input data from the user. Input data may include a message set including at least one message and a system response value for a dialog according to the message set. At least one message may be a text message using natural language that is or can be used in a dialog performed by the chatbot. The system response value to the dialog may be the chatbot's response value to any of at least one message or the chatbot's response value to the dialog.

After receiving input data, the computing device may determine a system function based on the input data. In some examples, the input data further includes a system function, and the computing device can determine the system function from the input data. In some other examples, the input data may not include a system function, and the computing device, after receiving the input data, may determine the system function based on at least one of a system response value and a message set in the input data. In some examples, a system function may be related to the type of dialog associated with the message set and may include one or more variables. In some other examples, system functions may be associated with the behavior of a chatbot. Process 400 may continue from block 410 to block 420, which determines one or more label data.

At block 420, the computing device may determine one or more label data for each of the at least one message based on the system function. Each label data may include data and at least one label value corresponding to the data. This label data is an element of learning data used for chatbot learning. In some examples, the computing device may derive one or more data from each message, determine a label value for each of the one or more derived data as predetermined by a system function, and associate the derived data with the label value. . In some other examples, the computing device may determine a predetermined label value required for a type of dialog, as predetermined by a system function, and obtain data corresponding to the determined label value from the message set and dialog-related data. there is. It may include at least one of an intent for a message, one or more entities, and a dialog state value within a dialog. The one or more label data is not limited to this and may include any data required as a variable in the learning data. Process 400 may continue from block 420 to block 430, which generates training data.

In block 430, the computing device may generate training data based on one or more label data determined in block 420. Learning data generated by a computing device may have a predetermined format so that it can be used as learning data used for chatbot learning.

In a further embodiment, process 400 may continue at block 440, which receives calibration training data from block 430. At block 440, the computing device may receive corrected training data from a user that corrects the training data generated by block 430. In some examples, the computing device may evaluate how the computing device determines its labeling value based on the training data generated and the received corrected training data at block 430 . Thereafter, the user may modify the method of determining the labeling value of the computing device, the system function, or the learning data based on the results of the evaluation. In some other examples, the computing device can adjust how the computing device determines one or more labeling values based on the training data and the corrected training data.

5 illustrates a computer program product 500 that can be used to perform generation of training data for a machine learning-based chatbot, according to at least some embodiments of the disclosure. . An exemplary embodiment of an exemplary computer program product is provided using signal bearing media 510. In some embodiments, signal bearing media 510 of one or more computer program products 500 may include computer-readable media 530 and/or recordable media 540.

Instructions 520 included in signal bearing medium 510 may be executed by a computing device, such as text-to-speech device 200 . When the instruction 520 is executed, it causes the computing device to include one or more instructions for receiving input data from a user, including a message set including at least one message and a system response value for a dialog according to the message set; For each of the at least one message, one or more commands for determining one or more label data to be used for training the chatbot, based on a system function related to the type of dialog; and at least one of one or more instructions for generating training data based on the one or more determined label data.

6 is a block diagram of an example embodiment of a computing device 600 arranged in accordance with at least some embodiments of the present disclosure. In one example basic configuration 602, computing device 600 may include one or more processors 604 and system memory 606. Memory bus 608 may be used to communicate between processor 604 and system memory 606.

Depending on the desired configuration, processor 604 may be of any type, including, but not limited to, a microprocessor (μP), microcontroller (μC), digital signal processor (DSP), or any combination thereof. The processor 604 may include one or more levels of cache, such as a level 1 cache 610 and a level 2 cache 612, a processor core 614, and a register 616. Processor core 614 may include an arithmetic logic unit (ALU), a floating point unit (FPU), a digital signal processing core (DSP Core), or any combination thereof. Memory controller 618 may also be used in conjunction with processor 604, or in some implementations, memory controller 618 may be an internal part of processor 604.

Depending on the desired configuration, system memory 606 may be of any type, including, but not limited to, volatile memory (such as RAM), non-volatile memory (such as ROM, flash memory, etc.), or any combination thereof. No. System memory 606 may include an operating system 620, one or more applications 622, and program data 624. Application 622 includes the functional blocks and/or actions described with respect to learning data generation device 200 of FIG. 2, chatbot system 300 of FIG. 3, and/or process 400 of FIG. 4, as described herein. It may include an algorithm 626 arranged to perform the function as described. Program data 624 may include data 628 for use with algorithm 626, such as data corresponding to a static network environment. In some embodiments, applications 622 may be arranged to operate with program data 624 on operating system 620 such that implementations for determining optimal transfer environments may be provided as described herein. For example, learning data generation device 200 may include all or part of computing device 600 and all or part of application 622 so that an implementation for determining an optimal transmission environment may be provided as described herein. It may be possible to do some of this. This described basic configuration is shown in Figure 6 by its components within dashed line 602.

Computing device 600 may have additional features or functionality, and additional interfaces, to facilitate communication between the basic configuration 602 and any desired devices and interfaces. For example, bus/interface controller 630 may be used to facilitate communication between basic configuration 602 and one or more data storage devices 632 via storage interface bus 634. Data storage device 632 may be a removable storage device 636, a non-removable storage device 638, or a combination thereof. Some examples of removable and non-removable storage devices include magnetic disk devices such as flexible disk drives and hard disk drives (HDD), compact disk (CD) drives, or digital versatile disk (DVD) drives, and Such as optical disk drives, solid state drives (SSD), and tape drives. Exemplary computer storage media include volatile and non-volatile removable and non-volatile media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. can do.

System memory 606, removable storage device 636, and non-removable storage device 638 are all examples of computer storage media. Computer storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage devices, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or Including, but not limited to, any other media that can be used to store required information and that can be accessed by computing device 600. Any such computer storage media may be part of computing device 600.

Computing device 600 also has an interface bus 642 to facilitate communication from various interface devices (e.g., output interfaces, peripheral interfaces, and communication interfaces) to base configuration 602 via bus/interface controller 642. ) may include. Exemplary output devices 642 may include graphics processing unit 648 and audio processing unit 650, which may be configured to communicate via one or more A/V ports 652 to various external devices, such as displays or speakers. It can be configured. Additionally, although graphics processing unit 648 is shown in Figure 6 as a component of an output device, it is possible to perform operations according to some implementations in addition to processor 604. According to some embodiments according to the present disclosure, the graphics processing unit 648 may be configured to perform various neural network-related operations, such as learning a neural network, processing according to a neural network, etc. Exemplary peripheral interface 644 may include a serial interface controller 654 or a parallel interface controller 656, which may provide input devices (e.g., keyboard, mouse, pen, etc.) through one or more I/O ports 658. It may be configured to communicate with external devices such as voice input devices, touch input devices, etc.) or other peripheral devices (e.g., printers, scanners, etc.). Exemplary communications device 646 includes a network controller 660, which may be arranged to facilitate communication with one or more other computing devices 662 over network communications via one or more communications ports 664.

Computing device 600 may also be implemented as a personal computer, including both laptop and non-laptop computer configurations. Additionally, computing device 600 may be implemented as part of a wireless base station or other wireless system or device.

FIG. 7 is an example of input data 700 to a learning data generating device, according to at least some embodiments of the present disclosure. The format of the input data 700 in FIG. 7 is merely an example for explanation, and those skilled in the art will understand that various changes can be made without departing from the spirit of the present disclosure. Example input data 700 in FIG. 7 relates to flight reservations. As shown in FIG. 7 , input data may include a message set 710 including at least one message such as message 712, a system function 720, and a system response value 730. The message set 710 may include a natural language message of a chatbot user such as “userUtterrance” and a natural language message of the chatbot such as “botUtterance”. The system function 720 includes not only the functions themselves for calling ("function call", "name", "checkSchedule"), but also functions included within such functions, such as ("departure": "Gimpo"), ("destination") : "Jeju"), ("start_date": "12th"), ("start_time": "9 o'clock"), and may contain one or more variables and corresponding values. The system response value 730 may include a value for the action performed by the chatbot according to the dialog. For example, as shown in FIG. 7, the system response value 730 may include ("flight_id": "A01234") and ("start_time": "9:10"), which are response values according to the search results of the aircraft. You can.

FIG. 8 is an example of learning data 800 generated based on the input data of FIG. 7. The format of the learning data 800 in FIG. 8 is merely an example for explanation, and those skilled in the art will understand that various changes can be made without departing from the spirit of the present disclosure. The learning data generating device according to the present disclosure may generate one or more label data 810 for the message 712 based on the system function 720 of FIG. 7, and in this manner, as shown in FIG. 8 Likewise, label data for each message set 710 can be generated, and training data 800 can be generated using the generated label data.

The description of the present application described above is for illustrative purposes, and those skilled in the art will understand that the present application can be easily modified into other specific forms without changing its technical idea or essential features. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. For example, each component described as single may be implemented in a distributed manner, and similarly, components described as distributed may also be implemented in a combined form.

Above, we have looked in detail at the object sought to be claimed in this disclosure. The subject matter claimed in this disclosure is not limited in scope to the specific implementation examples described above. For example, in some implementations it may be in the form of hardware operably used on a device or combination of devices, in other implementations it may be implemented in the form of software and/or firmware, and in still other implementations it may be implemented in the form of a signal bearing medium, It may contain one or more items such as storage media. Here, the storage medium, such as a CD-ROM, computer disk, flash memory, etc., may contain instructions that, when executed by a computing device, such as a computing system, computing platform, or other system, may cause execution of the corresponding processor according to the implementation described above. can be saved. Such a computing device may include one or more processing units or processors, one or more input/output devices such as a display, keyboard, and/or mouse, and one or more memory such as static random access memory, dynamic random access memory, flash memory, and/or a hard drive. may include.

The foregoing detailed description illustrates various embodiments of devices and/or processes through block diagrams, flow diagrams, and/or other examples. Such block diagrams, flowcharts, and/or other examples will include one or more functions and/or operations, and those of ordinary skill in the art will understand that each function and/or operation within the block diagrams, flowcharts, and/or other examples includes hardware, software, firmware, Alternatively, it will be understood that any combination thereof may be implemented individually or collectively. In one embodiment, some portions of the subject matter described in this disclosure may be implemented through an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA), Digital Signal Processor (DSP), or other form of integration. Alternatively, some aspects of embodiments of the present disclosure may include one or more computer programs running on one or more computers (e.g., one or more programs running on one or more computer systems), one or more programs running on one or more processors (e.g., one or more programs running on one or more computer systems), (e.g., one or more programs running on one or more microprocessors), firmware, or substantially any combination thereof, may also be implemented in whole or in part on an integrated circuit, including writing code for software and/or firmware. and/or design of circuits are within the scope of skill of those skilled in the art in light of this disclosure. Additionally, those skilled in the art will understand that the mechanisms of the subject matter of this disclosure may be distributed in a variety of program products, and the examples of the subject matter of this disclosure apply regardless of the specific type of signal bearing medium used to actually perform the distribution. You will understand.

Although certain example techniques have been described and shown herein using various methods and systems, those skilled in the art will recognize that various other modifications or equivalents may be substituted without departing from the claimed subject matter. Additionally, many modifications may be made to adapt the teachings of the claimed subject matter to particular situations without departing from the central concept described herein. Accordingly, it is intended that the claimed subject matter not be limited to the specific examples disclosed, but that such claimed subject matter may also include all embodiments that fall within the scope of the appended claims and their equivalents.

The scope of the present disclosure is indicated by the claims described below rather than the detailed description above, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present application. do.

Claims (13)

A learning data generation method performed under the control of a computing device and for generating learning data for a machine learning-based chatbot,
Receiving, from a user, input data including a message set including at least one message and a system response value for a dialog according to the message set;
determining, based on at least one of the message set and the system response value, a system function associated with the type of dialog and for associating data from each message with at least one label value associated with the type of dialog;
For each of the at least one message, determining one or more label data used for learning of the chatbot based on the system function; and
Generating learning data based on the determined one or more label data
A learning data generation method including. According to paragraph 1,
A method of generating learning data, wherein the system function is included in the input data. delete According to paragraph 1,
The method of generating learning data, wherein the step of determining the system function is further based on one or more predetermined learning data. According to paragraph 1,
The one or more label data for each of the at least one message includes at least one of an intent for the corresponding message, one or more entities, and a dialog state value within the dialog. Generating learning data. method. According to paragraph 1,
Receiving correction learning data that corrects the learning data from the user
A method of generating training data further comprising: According to clause 6,
Evaluating a method of determining a value in the one or more label data based on the learning data and the corrected learning data.
A method of generating training data further comprising: According to clause 6,
Adjusting a method of determining a value in the one or more label data based on the learning data and the corrected learning data.
A method of generating training data further comprising: According to paragraph 1,
A method of generating learning data, wherein each of the determined at least one label data includes data and at least one label value corresponding to the data. As a learning data generation device for a machine learning-based chatbot,
an input module that receives, from a user, input data including a message set including at least one message and a system response value for a dialog according to the message set;
Based on at least one of the message set and the system response value, determine a system function associated with the type of dialog and for associating data from each message with at least one label value associated with the type of dialog, a labeling processing module that determines, for each of at least one message, one or more label data used for learning of the chatbot, based on the system function; and
An output module that generates learning data based on the determined one or more label data
A learning data generating device comprising: According to clause 10,
A feedback processing module that receives correction learning data that corrects the learning data from the user
A learning data generating device further comprising: According to clause 11,
The feedback processing module is configured to evaluate the performance of the labeling processing module based on the learning data and the corrected learning data. A computer-readable storage medium that stores a labeling program that generates training data for a machine learning-based chatbot, where the labeling program is executed by a computing device,
Receiving, from a user, input data including a message set including at least one message and a system response value for a dialog according to the message set;
determining, based on at least one of the message set and the system response value, a system function associated with the type of dialog and for associating data from each message with at least one label value associated with the type of dialog;
For each of the at least one message, determining one or more label data used for learning of the chatbot based on the system function; and
An operation of generating learning data based on the determined one or more label data
A computer-readable storage medium comprising one or more instructions to perform.