KR102857236B1 - Electronic apparatus and method for controlling thereof - Google Patents

Abstract

An electronic device is disclosed. The electronic device includes a memory storing computer executable instructions and a processor that executes the computer executable instructions to determine a text corresponding to a received user command, and if the text is classified as corresponding to one of a plurality of pre-stored texts through a first artificial intelligence model, provides response information for the user command, and if the text is classified as an error through the first artificial intelligence model, provides error information for the user command, wherein the first artificial intelligence model classifies the text as an error if the text corresponding to the user command is a similar text that differs in one of an entity and an intent from at least one of the plurality of pre-stored texts.

Description

Electronic apparatus and method for controlling it {ELECTRONIC APPARATUS AND METHOD FOR CONTROLLING THEREOF}

The present disclosure relates to an electronic device and a method for controlling the same, and more particularly, to an electronic device operating based on artificial intelligence technology and a method for controlling the same.

Artificial intelligence systems are being developed recently. Unlike conventional rule-based systems, AI systems are machines that learn and make decisions on their own. They are being utilized in various technological fields, such as voice recognition, image recognition, and future prediction.

In particular, conversational AI systems are being developed recently. These systems analyze user requests through AI models and provide responses. They are being utilized in services such as chatbots.

Typically, conversational AI systems store vast amounts of text in a database. When a user requests a request, they search for the text most relevant to the request among the texts stored in the database and provide a response based on that text.

However, in some cases, users may enter requests that are similar in form to the text stored in the database, but have different topics, purposes, etc.

For example, if the text stored in the database is 'Find a taxi' in the 'Transportation' category, and the user enters 'I want to watch Taxi' with the intention of watching the movie Taxi, this is the case.

In this case, if there is no text such as 'I want to appreciate taxis' in the database, the conventional conversational artificial intelligence system finds 'Find a taxi', which is most similar to 'I want to appreciate taxis' among the stored texts, and provides information related to available taxis around the user as response information.

However, this is response information that has nothing to do with the user's intention and only causes inconvenience to the user.

The present disclosure has been devised to solve the above-described problems, and the purpose of the present disclosure is to provide an electronic device that can distinguish text that is similar in form to a text pre-stored in a database but has a different subject, purpose, etc. from the text pre-stored in the database.

According to an embodiment of the present disclosure for achieving the above object, an electronic device includes a memory for storing computer executable instructions and a processor for determining a text corresponding to a received user command by executing the computer executable instructions, and providing response information for the user command when the text is classified as corresponding to one of a plurality of pre-stored texts through a first artificial intelligence model, and providing error information for the user command when the text is classified as an error through the first artificial intelligence model, wherein the first artificial intelligence model can classify the text as an error when the text corresponding to the user command is a similar text that differs in one of an entity and an intent from at least one of the plurality of pre-stored texts.

Here, the first artificial intelligence model may be a model obtained by training a model that determines a text corresponding to the user command based on the plurality of pre-stored texts to classify the similar text as an error.

And, the above similar text can be generated by the second artificial intelligence model.

And, the first artificial intelligence model can determine whether the similar text input to the first artificial intelligence model corresponds to one of the plurality of pre-stored texts, and determine whether to learn based on the determination result.

In addition, the first artificial intelligence model may learn to classify the similar text as an error if it is determined that the similar text corresponds to one of the plurality of pre-stored texts, and may not perform the learning if it is determined that the similar text does not correspond to one of the plurality of pre-stored texts.

And, the second artificial intelligence model can learn to generate a new similar text having a higher degree of similarity than the degree to which the similar text is similar to at least one text, if the first artificial intelligence model determines that the similar text does not correspond to the plurality of pre-stored texts.

And, the first artificial intelligence model can receive the new similar text generated by the second artificial intelligence model as input, determine whether the new similar text corresponds to one of the plurality of pre-stored texts, and determine whether to learn based on the determination result.

And, the first and second artificial intelligence models may be GAN (Generative Adversarial Network).

Meanwhile, a method for controlling an electronic device according to an embodiment of the present disclosure includes a step of determining a text corresponding to a received user command, and a step of providing response information for the user command if the text is classified as corresponding to one of a plurality of pre-stored texts through a first artificial intelligence model, and a step of providing error information for the user command if the text is classified as an error through the first artificial intelligence model, wherein the first artificial intelligence model can classify the text as the error if the text corresponding to the user command is a similar text that differs in one of an entity and an intent from at least one of the plurality of pre-stored texts.

Here, the first artificial intelligence model may be a model obtained by training a model that determines a text corresponding to the user command based on the plurality of pre-stored texts to classify the similar text as an error.

And, the above similar text can be generated by the second artificial intelligence model.

And, the first artificial intelligence model can determine whether the similar text input to the first artificial intelligence model corresponds to one of the plurality of pre-stored texts, and determine whether to learn based on the determination result.

In addition, the first artificial intelligence model may learn to classify the similar text as an error if it is determined that the similar text corresponds to one of the plurality of pre-stored texts, and may not perform the learning if it is determined that the similar text does not correspond to one of the plurality of pre-stored texts.

And, the second artificial intelligence model can learn to generate a new similar text having a higher degree of similarity than the degree to which the similar text is similar to at least one text, if the first artificial intelligence model determines that the similar text does not correspond to the plurality of pre-stored texts.

And, the first artificial intelligence model can receive the new similar text generated by the second artificial intelligence model as input, determine whether the new similar text corresponds to one of the plurality of pre-stored texts, and determine whether to learn based on the determination result.

And, the first and second artificial intelligence models may be GAN (Generative Adversarial Network).

According to various embodiments of the present disclosure as described above, an electronic device can be provided that can distinguish a text that is morphologically similar to a text pre-stored in a memory but has a different entity or intent from the text pre-stored in the memory.

Accordingly, it is possible to prevent cases where response information that is completely unrelated to the user's intention is provided.

FIG. 1 is a drawing for explaining a screen provided by an electronic device according to one embodiment of the present disclosure.
FIG. 2 is a block diagram illustrating an electronic device according to an embodiment of the present disclosure.
FIG. 3 is a diagram illustrating a plurality of texts stored in a memory according to an embodiment of the present disclosure.
FIG. 4 is a drawing for explaining additionally stored text according to one embodiment of the present disclosure.
FIG. 5 is a diagram illustrating learning performed by an artificial intelligence model according to an embodiment of the present disclosure.
FIGS. 6A to 9B are diagrams illustrating various embodiments of response information according to an embodiment of the present disclosure.
FIG. 10 is a detailed block diagram of an electronic device implemented as a server according to an embodiment of the present disclosure.
FIG. 11 is a detailed block diagram of an electronic device according to an embodiment of the present disclosure when implemented as a display device.
FIG. 12 is a drawing for explaining a conversation system according to one embodiment of the present disclosure.
FIG. 13 is a detailed block diagram of an electronic device according to an embodiment of the present disclosure.
FIG. 14 is a flowchart illustrating the operation of an electronic device according to an embodiment of the present disclosure.

First, the terms used in this specification and claims are general terms selected in consideration of the functions of the present disclosure. However, these terms may vary depending on the intentions of those skilled in the art, legal or technical interpretations, and the emergence of new technologies. Furthermore, some terms may have been arbitrarily selected by the applicant. These terms may be interpreted according to the meanings defined in this specification. In the absence of a specific definition, they may be interpreted based on the overall content of this specification and common technical knowledge in the relevant technical field.

In addition, when describing the present disclosure, if it is determined that a specific description of a related known function or configuration may unnecessarily obscure the gist of the present disclosure, the detailed description thereof will be abbreviated or omitted.

Furthermore, embodiments of the present disclosure will be described in detail with reference to the attached drawings and the contents described in the attached drawings, but the present disclosure is not limited or restricted by the embodiments.

Hereinafter, the present disclosure will be described in detail with reference to the attached drawings.

FIG. 1 is a drawing for explaining a screen provided by an electronic device according to one embodiment of the present disclosure.

The electronic device (100) may be any of a variety of electronic devices capable of receiving user commands. For example, the electronic device (100) may be a smart phone, a PC, a tablet PC, a laptop, a PDA, a PMP, a smart TV, etc.

Additionally, the electronic device (100) may be a server that receives user commands from an external device (not shown). When the electronic device (100) is implemented as a server, the external device (not shown) may be a smart phone, a PC, a tablet PC, a laptop, a PDA, a PMP, a smart TV, etc. In the following, for convenience of explanation, the electronic device (100) according to one embodiment of the present disclosure is assumed to be a server.

The electronic device (100) may provide a screen that prompts a user's inquiry. For example, as illustrated in (a) of FIG. 1, the electronic device (100) may provide a screen containing a message such as "How can I help you?" Here, the screen may be the execution screen of a specific application.

In addition, the electronic device (100) can receive a user command. Here, the user command can be text entered through a keyboard or an on-screen keyboard, or a picture or text entered through a stylus or the like. In addition, the user command can also be the user's voice entered through a microphone.

And, when a user command is received, the electronic device (100) can provide response information to the user command.

To this end, the electronic device (100) can first classify the category of the user command. Specifically, when a user command is received, the electronic device (100) can identify a text corresponding to the user command from among a plurality of pre-stored texts, and classify the category of the user command based on the category information matched to each text. In addition, the electronic device (100) can provide information corresponding to the user command from among a plurality of pieces of information included in the corresponding category as response information.

For example, when a text such as 'taxi search' is pre-stored in the electronic device (100), and "find a taxi" is input as illustrated in (b) of FIG. 1, the electronic device (100) can identify 'taxi search' as the text corresponding to "find a taxi" among the plurality of pre-stored texts. In addition, the electronic device (100) can classify the 'transportation' category matching 'taxi search' as a category of the user command, and provide information related to 'taxi search' among the plurality of pieces of information included in the 'transportation' category as response information. That is, as illustrated in (b) of FIG. 1, the electronic device (100) can provide information related to taxis available for reservation around the user as response information.

Meanwhile, an electronic device (100) according to an embodiment of the present disclosure may classify the category of the user command as an error if the text corresponding to the received user command is similar in form to a plurality of pre-stored texts, but one of the entity and intent is different.

For example, in a case where a text such as 'taxi search' is pre-stored in the electronic device (100) as in the above-described embodiment, when "I want to see a taxi" is input as shown in (c) of FIG. 1, 'taxi search' and "I want to see a taxi" are similar in form in that they both contain 'taxi', but the former is intended to search for information related to 'transportation' taxis, and the latter is intended to search for information related to 'movie' taxis, and thus the two intentions are different. In this case, the electronic device (100) may classify the category of the corresponding user command as an error. A detailed description thereof will be provided later.

In addition, the electronic device (100) can provide error information when the category of the user command is classified as an error. Here, the error information can be information indicating that the information requested by the user command cannot be found.

Referring again to (c) of FIG. 1, when text such as ‘taxi search’ is pre-stored in the electronic device (100), and “I want to see a taxi” is entered, the electronic device (100) can classify the category of “I want to see a taxi” as an error and provide information that the requested information cannot be found as error information.

Accordingly, unlike conventional electronic devices, the electronic device (100) according to one embodiment of the present disclosure can prevent cases where response information that is completely unrelated to the user's intention is provided.

FIG. 2 is a block diagram illustrating an electronic device according to an embodiment of the present disclosure. As illustrated in FIG. 2, the electronic device (100) includes a memory (110) and a processor (120).

The memory (110) may include, for example, built-in memory or external memory. The built-in memory may include, for example, at least one of volatile memory (e.g., dynamic RAM (DRAM), static RAM (SRAM), or synchronous dynamic RAM (SDRAM)), non-volatile memory (e.g., one time programmable ROM (OTPROM), programmable ROM (PROM), erasable and programmable ROM (EPROM), electrically erasable and programmable ROM (EEPROM), mask ROM, flash ROM, flash memory (e.g., NAND flash or NOR flash), a hard drive, or a solid state drive (SSD).

External memory may include a flash drive, such as a compact flash (CF), secure digital (SD), micro secure digital (Micro-SD), mini secure digital (Mini-SD), extreme digital (xD), multi-media card (MMC), or memory stick.

The memory (110) is accessed by the processor (120), and data can be read/written/modified/deleted/updated by the processor (120).

In the present disclosure, the term memory may include at least one of a memory provided separately from the processor (120), a ROM (not shown) and a RAM (not shown) within the processor (120).

The processor (120) is a component for controlling the overall operation of the electronic device (100). For example, the processor (120) may control a plurality of hardware or software components connected to the processor (120) by running an operating system or an application program, and may perform various data processing and calculations. The processor (120) may be a central processing unit (CPU), a graphics-processing unit (GPU), or both. The processor (120) may be implemented as at least one general processor, a digital signal processor, an application specific integrated circuit (ASIC), a system on chip (SoC), a microcomputer (MICOM), etc.

The processor (120) can perform operations of the electronic device (100) according to various embodiments of the present disclosure by executing computer executable instructions stored in the memory (110).

For example, the processor (120) can provide response information to a user command by executing computer executable instructions stored in the memory (110). This will be described in detail below with reference to FIGS. 3 to 5.

FIG. 3 is a diagram illustrating a plurality of texts stored in a memory according to an embodiment of the present disclosure.

Referring to FIG. 3, the memory (110) can store multiple texts. Here, each of the multiple texts may be associated with information regarding a category. For example, Text 1 may be associated with Label 1, and Text 2 may be associated with Label 2. As a more specific example, the "Transportation" category may be associated with "Taxi Search," and the "Broadcast" category may be associated with "Cooking Broadcast."

The processor (120) can receive a user command. Here, the user command can be text inputted through a keyboard or an on-screen keyboard of an external device (not shown). Additionally, the user command can be a user voice inputted through a microphone of the external device (not shown). In this case, the processor (120) can convert the received user voice into text through an STT (Speech to Text) algorithm. For convenience of explanation, the following description assumes that the user command is text inputted through an on-screen keyboard of an external device (not shown).

The processor (120) can identify a text corresponding to a user command among a plurality of pre-stored texts.

Specifically, when a user command is received, the processor (120) can identify a text corresponding to the user command through natural language processing technology. Here, the natural language processing can be performed using a first artificial intelligence model. Specifically, the processor (120) can analyze the morphology, syntax, structure, etc. of the user command using the first artificial intelligence model, and identify a text corresponding to the user command among a plurality of pre-stored texts that matches the received user command by a preset threshold value or more.

For example, if the user command is “Find me a taxi,” the processor (120) may use the first artificial intelligence model to analyze the morphology, syntax, structure, etc. of “Find me a taxi,” and identify a text corresponding to the user command among a plurality of pre-stored texts that matches “Find me a taxi” by a preset threshold value or more. In the above-described embodiment, the text corresponding to “Find me a taxi” may be “Search for a taxi.”

Thereafter, the processor (120) can classify the categories of the user commands. Specifically, the processor (120) can classify the categories of the user commands using category information matched to each text.

For example, if text 1 in FIG. 3 is 'taxi search' and Label 1 is 'transportation' category, when the processor (120) receives a user command of "find a taxi", it can classify the category of the user command as 'transportation'.

In addition, the processor (120) can provide information corresponding to the user command as response information among a plurality of pieces of information included in the category of the user command.

In the above-described embodiment, the processor (120) may provide information related to "taxi search" from among the plurality of pieces of information included in the "transportation" category as response information. In this case, the response information may be information related to available taxis around the user. To this end, the processor (120) may receive location information from an external device (e.g., a smartphone) into which a user command has been input, or receive taxi-related information from an external server that provides information on traffic conditions.

On the other hand, if response information is provided solely based on text already stored in memory, information unrelated to the user's intent may be provided as response information. For example, if a user enters "I want to see Taxi" with the intention of watching the movie Taxi, but no text related to the movie is stored in memory, this could occur.

In this case, the conventional electronic device retrieved the text "Find a taxi" from the pre-stored texts, which most closely resembled "I want to see a taxi." In response, it provided information about available taxis in the user's vicinity. This information is completely unrelated to the user's intention to watch the movie "Taxi," and can be inconvenient.

FIG. 4 is a drawing for explaining additionally stored text according to one embodiment of the present disclosure.

Referring to FIG. 4, the memory (110) can store a new text (text N+1) in addition to a plurality of pre-stored texts (text 1 to text N). Here, the new text may be a text that is morphologically similar to the plurality of pre-stored texts, but has a different entity or intent.

Specifically, the new text may be a text that matches at least one of the multiple pre-stored texts with respect to one of the target and intent, but does not match the other.

For example, if the pre-saved text is "Search for a Taxi" and matches the "Transportation" category, a new text, such as "Appreciate a Taxi," could match the target as "taxi" but have a different intent. Furthermore, if the pre-saved text is "Turn up the volume" and matches the "Volume" category, a new text, such as "Turn up the channel," could match the target, but have the same intent of increasing a specific value.

Meanwhile, the new text (N+1) may be associated with an error category. Here, the error category can be expressed by various names, such as the Rejection category.

The processor (120) can identify a text corresponding to a user command among a plurality of pre-stored texts.

As described above, when a user command is received, the processor (120) can identify text corresponding to the user command through natural language processing technology.

And, if the text corresponding to the user command is identified as one of the plurality of pre-stored texts (text 1 to text N), the processor (120) can provide information corresponding to the user command as response information among the plurality of pieces of information included in the category of the user command.

If the text corresponding to the user command is identified as one of the generated texts, the processor (120) may provide error information indicating that the information requested by the user command cannot be found. In other words, the processor (120) may reject the user command.

Specifically, if the text corresponding to the user command is identified as one of the new texts, the processor (120) can classify the category of the user command as an error and reject the user command based on the category information matched to the new text.

Meanwhile, new texts like this can be generated by the second artificial intelligence model and classified as errors by the first artificial intelligence model.

FIG. 5 is a diagram illustrating learning performed by an artificial intelligence model according to an embodiment of the present disclosure.

The second artificial intelligence model (520) can generate new texts that are different from the previously stored multiple texts. For convenience of explanation, the new texts generated by the second artificial intelligence model (520) will be referred to as "similar texts" below.

Specifically, the second artificial intelligence model (520) can take a plurality of pre-stored texts as input and generate a text that is morphologically similar to the plurality of pre-stored texts, but has a different entity or intent.

For example, if the pre-stored text is 'taxi search' that matches the 'transportation' category, the second artificial intelligence model (520) can generate similar texts such as 'I want to see a taxi' and 'appreciate a taxi' that are similar in form to 'taxi search' but have different intentions. In addition, if the pre-stored text is 'turn up the volume' that matches the 'volume' category, the second artificial intelligence model (520) can generate similar texts such as 'turn down the volume' that are similar in form to 'turn up the volume' but have different targets and intentions.

The first artificial intelligence model (510) can input a similar text generated by the second artificial intelligence model (520) and determine whether the similar text is one of multiple pre-stored texts. Furthermore, if the first artificial intelligence model (510) determines that the similar text is one of multiple pre-stored texts, it can learn to classify the similar text as an error.

Specifically, the first artificial intelligence model (510) can be trained to classify a similar text as an error if it is determined to be one of the plurality of pre-stored texts by taking as input a plurality of pre-stored texts and a similar text generated by the second artificial intelligence model (520).

For example, if the text "taxi search" is pre-stored in the "transportation" category, but a text such as "taxi appreciation" is not pre-stored, and "find a taxi" is generated as a similar text by the second artificial intelligence model, the first artificial intelligence model (510) can determine whether the text corresponding to "find a taxi" is one of the multiple pre-stored texts. In addition, if the first artificial intelligence model (510) determines that the pre-stored "taxi search" is content corresponding to the similar text "find a taxi," the first artificial intelligence model (510) can learn to classify the similar text as an error.

Here, learning can be weight adjustment between neurons included in different layers of the first artificial intelligence model (501). Specifically, the first artificial intelligence model can adjust the weight between neurons included in different layers so as to output a uniform output vector value (uniform distribution) for similar texts generated by the second artificial intelligence model (520). For example, if N texts are pre-stored, the first artificial intelligence model (510) can adjust the weight between neurons so as to output 1/N for similar texts generated by the second artificial intelligence model (520).

That is, the first artificial intelligence model (501) can be trained to output 1/N for similar texts generated by the second artificial intelligence model (520). Accordingly, the first artificial intelligence model (501) can classify a user command as an error if the output vector value for a subsequently received user command is output as 1/N.

Meanwhile, if the first artificial intelligence model (510) classifies the similar text generated by the second artificial intelligence model (520) as an error, the first artificial intelligence model (510) may not perform the learning described above. That is, if the first artificial intelligence model (510) determines that the similar text does not correspond to one of the plurality of pre-stored texts, the first artificial intelligence model (510) may not perform the learning described above.

In this case, a second artificial intelligence model (520) may be trained. Specifically, if the first artificial intelligence model (510) classifies a similar text as an error, the second artificial intelligence model (520) may be trained to generate a text that is more similar to at least one of the pre-stored texts than the similar text described above.

Here, learning may be weight adjustment between neurons included in different layers of the second artificial intelligence model (502). Specifically, the second artificial intelligence model may adjust the weights between neurons included in different layers to generate text that is more similar to at least one of the pre-stored texts than the aforementioned similar text.

For example, if the pre-stored text is 'volume up' that matches the 'volume' category, the second artificial intelligence model (520) can generate similar texts such as 'volume down' that are similar in form to 'volume up' but have different targets and intentions.

In addition, if the first artificial intelligence model (510) classifies the similar text 'volume down' as an error, the second artificial intelligence model (520) can generate a text more similar to 'volume up' than 'volume down' through learning. For example, the newly generated similar text could be 'don't turn up the volume'.

Thereafter, the first artificial intelligence model (510) can input a new similar text having a higher similarity than the similar text generated by the second artificial intelligence model (502), and determine whether the new similar text is one of the plurality of previously stored texts.

In addition, if the first artificial intelligence model (501) determines that the new similar text is one of the multiple pre-stored texts, the first artificial intelligence model (501) can perform the above-described learning again. Similarly, if the first artificial intelligence model (501) classifies the new similar text as an error, the second artificial intelligence model (502) can perform the above-described learning again.

Through this kind of repetitive learning, the first artificial intelligence model (501) can classify similar texts generated by the second artificial intelligence model (501) as errors.

Meanwhile, the artificial intelligence model of the present disclosure described above may be a Generative Adversarial Network (GAN) model.

A GAN is a model that generates fakes that look real through competition between two neural network models. The second artificial intelligence model (502) can be a generator model of the GAN. Specifically, the second artificial intelligence model (502) can take multiple pre-stored texts as input and generate texts that are morphologically similar to the multiple pre-stored texts, but have at least one target and/or intent. In other words, the purpose of the second artificial intelligence model (502) is to generate texts similar to multiple pre-stored lists.

The first artificial intelligence model (501) can be a discriminator model of a GAN. Specifically, the first artificial intelligence model (501) can determine whether the text generated by the second artificial intelligence model (502) is one of multiple pre-stored texts or not.

And, the first artificial intelligence model (501) performs learning by inputting a text that is judged to be one of a plurality of pre-stored texts among the texts generated by the second artificial intelligence model (502), and can classify the text generated by the second artificial intelligence model (502) as an error through learning.

In addition, the second artificial intelligence model (502) performs learning by inputting text that the first artificial intelligence model (501) determines is not one of the plurality of pre-stored texts among the generated similar texts, and through learning, can generate text that has a higher similarity with the plurality of pre-stored texts than the previously generated similar texts.

Through repetition of such learning, the second artificial intelligence model (502) can generate texts having a high similarity to multiple pre-stored texts, and the first artificial intelligence model (501) can classify texts having a high similarity to multiple pre-stored texts as errors.

FIGS. 6A to 9B are diagrams illustrating various embodiments of response information according to an embodiment of the present disclosure. In the following description, it is assumed that "Find cooking shows" is pre-stored in the "Cooking" category, "Find taxis" is pre-stored in the "Transportation" category, and the first artificial intelligence model has been trained to classify "Find cooking ingredients" and "I want to see a taxi" into the "Error" category.

Figures 6a to 7b are drawings for explaining an embodiment in which a user's voice is input to a smart TV.

The electronic device (100) can receive user commands from an external device. For example, as illustrated in FIG. 6A, if a user utters "Find a cooking show" to a smart TV, which is an external device, the electronic device (100) can receive a voice signal corresponding to "Find a cooking show" from the external device.

In addition, the electronic device (100) can convert the voice signal into text by applying an STT (Speech to Text) algorithm to the voice signal.

In addition, the electronic device (100) can identify a text corresponding to a user command from among a plurality of pre-stored texts. Specifically, the electronic device (100) can identify a text corresponding to a user command from among a plurality of pre-stored texts using a first artificial intelligence model.

Additionally, the electronic device (100) can provide response information to a user command based on category information matched to the identified text.

For example, as illustrated in FIG. 6A, when the user command is “Find a cooking show,” the electronic device (100) can identify a text (e.g., “cooking show”) corresponding to “Find a cooking show” among a plurality of pre-stored texts, and provide response information for the user command based on category information (e.g., “cooking”) matched to the identified text.

That is, in the above-described embodiment, the electronic device (100) may provide information related to 'cooking broadcast' among a plurality of pieces of information within the 'cooking' category as response information, as illustrated in FIG. 6b. Meanwhile, the response information may be one of the pieces of information pre-stored in the electronic device (100), or may be information searched on the web based on keywords in a user command.

Meanwhile, as illustrated in FIG. 7a, when a user utters “Find cooking ingredients” to an external device, such as a smart TV, the electronic device (100) can receive a voice signal corresponding to “Find cooking ingredients” from the external device.

And, the electronic device (100) can identify a text corresponding to a user command among a plurality of pre-stored texts using the first artificial intelligence model.

Additionally, the electronic device (100) can provide error information for a user command based on category information matched to the identified text.

For example, as illustrated in FIG. 7a, when the user command is “Find cooking ingredients,” the electronic device (100) can identify a text corresponding to “Find cooking ingredients” (e.g., “Find cooking ingredients”) among a plurality of pre-stored texts, and provide error information for the user command based on category information (e.g., “Error”) matched to the identified text.

In the above-described embodiment, the electronic device (100) may classify the user command as an error and provide a message indicating that a search result for the user command cannot be found as error information, as illustrated in FIG. 7b.

FIGS. 8A to 9B are drawings for explaining an embodiment in which a user command is input to a user terminal device.

The electronic device (100) can receive user commands from an external device. For example, as illustrated in FIG. 8A, if a user inputs "Find a taxi" into a user terminal device, which is an external device, the electronic device (100) can receive a text such as "Find a taxi" from the external device. Here, the screen provided by the user terminal device can be the execution screen of the chatbot application.

In addition, the electronic device (100) can identify a text corresponding to a user command from among a plurality of pre-stored texts. Specifically, the electronic device (100) can identify a text corresponding to a user command from among a plurality of pre-stored texts using a first artificial intelligence model.

Additionally, the electronic device (100) can provide response information to a user command based on category information matched to the identified text.

For example, as illustrated in FIG. 8A, when the user command is “Find a taxi,” the electronic device (100) can identify a text corresponding to “Find a taxi” (e.g., “Find a taxi,” “Search for a taxi,” etc.) from among a plurality of pre-stored texts, and provide response information for the user command based on category information (e.g., “Transportation,” etc.) matched to the identified text.

In the above-described embodiment, the electronic device (100) may provide information related to "reservable taxis" among a plurality of pieces of information within the "transportation" category as response information, as illustrated in FIG. 8B . Meanwhile, the electronic device (100) may receive a GPS signal from a user terminal device or communicate with an external server that provides information on traffic conditions to search for reservable taxis around the user.

Meanwhile, as illustrated in FIG. 9a, when a user inputs “I want to see a taxi” into a user terminal device, which is an external device, the electronic device (100) can receive a text such as “I want to see a taxi” from the external device.

And, the electronic device (100) can identify a text corresponding to a user command among a plurality of pre-stored texts using the first artificial intelligence model.

Additionally, the electronic device (100) can provide error information for a user command based on category information matched to the identified text.

For example, as illustrated in FIG. 9A, when the user command is “I want to see a taxi,” the electronic device (100) can identify a text corresponding to “I want to see a taxi” (e.g., “I want to see a taxi”) from among a plurality of pre-stored texts, and provide error information for the user command based on category information (e.g., “error”) matched to the identified text.

In the above-described embodiment, the electronic device (100) may classify the user command as an error and provide a message as response information that induces a new user command along with a message that the search result for the user command cannot be found, as illustrated in FIG. 9b.

FIG. 10 is a detailed block diagram of an electronic device implemented as a server according to an embodiment of the present disclosure.

When the electronic device (100) is implemented as a server, the electronic device (100) may include a memory (1010), a communication unit (1020), and a processor (1030), as illustrated in FIG. 10. In the following, any part that overlaps with the above description will be omitted or abbreviated.

The memory (1010) can store commands or data related to at least one other component of the electronic device (100). In particular, the memory (1010) can be implemented as a non-volatile memory, a volatile memory, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or the like.

The memory (1010) is accessed by the processor (1030), and data reading/recording/modifying/deleting/updating, etc. can be performed by the processor (1030). In the present disclosure, the term "memory" may include a memory (1010), a ROM (not shown), a RAM (not shown) within the processor (1030), or a memory card (not shown) (e.g., a micro SD card, a memory stick) mounted on the electronic device (100). In addition, a plurality of texts may be pre-stored in the memory (1010), and each of the plurality of texts may be matched with a category.

Additionally, the memory (1010) may store a second artificial intelligence model that generates a text that is morphologically similar to a plurality of pre-stored texts but has a different entity and intent, and a first artificial intelligence model that determines whether a text generated by the second artificial intelligence model is one of the plurality of pre-stored texts.

The communication unit (1020) can communicate with an external device. At this time, the communication unit (1020) can receive a user command from the external device and transmit response information for the user command to the external device.

For this purpose, the communication unit (1020) may include a wireless communication chip, a Wi-Fi chip, Bluetooth, etc.

The processor (1030) is electrically connected to the memory (1010) and the communication unit (1020) to control the overall operation and function of the electronic device (100).

The processor (1030) can receive user commands from an external device through the communication unit (1020).

In addition, the processor (1030) determines a text corresponding to a received user command, and if the text is classified as corresponding to one of a plurality of pre-stored texts through the first artificial intelligence model, response information for the user command can be provided.

Specifically, the processor (1030) may determine, through the first artificial intelligence model, whether a text corresponding to a user command corresponds to one of a plurality of pre-stored texts. Furthermore, if the text is classified as corresponding to one of the plurality of pre-stored texts, the processor (1030) may determine a category matching the text and provide information corresponding to the user command from among a plurality of pieces of information included in the corresponding category as response information.

Here, the response information can be status information such as traffic conditions, as well as control information for turning on and off the power of an external device such as a TV.

Additionally, the processor (1030) can provide error information for the user command if the text corresponding to the user command is classified as an error through the first artificial intelligence model.

Specifically, the first artificial intelligence model may classify the text corresponding to the user command as an error if the text corresponding to the user command is similar to at least one of the plurality of pre-stored texts in that the text differs in one of the entity and intent. In this case, the processor (1030) may provide error information regarding the user command. Here, the error information may be information indicating that the requested information by the user command cannot be found.

FIG. 11 is a detailed block diagram of an electronic device according to an embodiment of the present disclosure when implemented as a display device.

The electronic device (100) may be implemented as a display device into which user commands can be input. For various examples, the electronic device (100) may be a smart phone, a PC, a tablet PC, a laptop, a PDA, a PMP, a smart TV, etc.

As illustrated in FIG. 11, when the electronic device is implemented as a display device, the electronic device (100) may include a memory (1110), an input unit (1130), a display (1140), a communication unit (1150), and a processor (1160).

The input unit (1130) can receive user commands.

The input unit (1130) may include a touch sensor, a (digital) pen sensor, a pressure sensor, or a key. The touch sensor may use, for example, at least one of a capacitive type, a pressure-sensitive type, an infrared type, or an ultrasonic type. The (digital) pen sensor may be, for example, part of a touch panel or may include a separate recognition sheet. The key may include, for example, a physical button, an optical key, or a keypad.

Additionally, the input unit (1130) may include a microphone. Here, the microphone may receive a user's spoken voice.

The display (1140) can display various screens. In particular, the display (1140) can provide a screen that prompts a user command. Additionally, the display (1140) can display response information to the user command.

The display (1140) may be implemented with various display technologies such as LCD (Liquid Crystal Display), OLED (Organic Light Emitting Diodes), AM-OLED (Active-Matrix Organic Light-Emitting Diode), LcoS (Liquid Crystal on Silicon), or DLP (Digital Light Processing). In addition, the display (1140) may be combined with at least one of the front area, side area, and rear area of the electronic device (100) in the form of a flexible display.

Additionally, the display (1140) may be implemented as a touch screen equipped with a touch sensor.

The communication unit (1150) can communicate with an external device. In particular, the communication unit (1150) can receive search results for text corresponding to a user command through communication with an external server. Furthermore, if the user command is a command for controlling a specific external device, the communication unit (1150) can transmit a control signal to the external device.

The processor (1160) can receive a user command through the input unit (1130). In addition, the processor (1160) can identify a text corresponding to the user command among a plurality of pre-stored texts using the first artificial intelligence model.

In addition, the processor (1160) can classify the category of the user command using the first artificial intelligence model and provide response information for the user command.

Specifically, the processor (1160) can display response information such as traffic conditions through the display (1140), or transmit a control signal to an external device such as a TV to turn the power of the external device on or off.

Meanwhile, when the electronic device (100) is implemented as a display device, some of the above-described technologies may be implemented in the display device, and the remaining parts may be implemented in the server.

Specifically, the display device receives a user command and transmits it to a server, the server obtains or generates response information for the received user command and transmits it to the display device, and the display device can display the received response information.

FIG. 12 is a drawing for explaining a conversation system according to one embodiment of the present disclosure.

An electronic device (100) according to an embodiment of the present disclosure may store a dialogue system that provides a response to a user input (particularly, a user voice). At this time, the dialogue system may include an Automatic Speech Recognition (ASR) unit (1210), a Natural Language Understanding (NLU) unit (1220), a Dialogue Manager (DM) (1230), a Natural Language Generator (NLG) unit (1240), a Text-to-Speech (TTS) unit (1250), and a Knowledge Database (1260), as illustrated in FIG. 12 .

The automatic speech recognition unit (1210) can perform speech recognition on a user's speech input through a microphone, etc. The natural language understanding unit (1220) can identify the intent of the user's speech based on the speech recognition result. The conversation manager (1230) can obtain information on a response to the user's speech based on the natural language understanding result and data stored in the knowledge database (1260). For example, the conversation manager (1230) can obtain information for generating a response, and as described above, the obtained information can be determined based on the intent of the user's speech identified through the natural language understanding unit (1220) and data stored in the knowledge database (1260). The natural language generation unit (1240) can obtain natural language as a response to the user's speech based on the information obtained through the conversation manager (1230). TTS (1250) can convert acquired natural language into voice. By doing so, the conversation system can provide a response to the user's voice as voice, so that the user can conduct a conversation with the electronic device (100).

In particular, a conversation manager (1230) according to one embodiment of the present disclosure can obtain response information for a user's voice based on a category matched to each of a plurality of texts stored in a knowledge database (1260).

FIG. 13 is a detailed block diagram of an electronic device according to an embodiment of the present disclosure.

As illustrated in FIG. 13, the electronic device (100) may include at least one of an image acquisition unit (1310), an image processing unit (1320), a display (1330), a communication unit (1340), a memory (1350), an audio processing unit (1360), an audio output unit (1370), an input unit (1380), and a processor (1390). Meanwhile, the configuration of the electronic device (100) illustrated in FIG. 13 is merely an example and is not necessarily limited to the block diagram described above. Accordingly, it goes without saying that some of the configurations of the electronic device (100) illustrated in FIG. 13 may be omitted, modified, or added depending on the type of the electronic device (100) or the purpose of the electronic device (100).

The image acquisition unit (1310) can acquire image data from various sources. For example, the image acquisition unit (1310) can acquire a specific image as response information to a user command. The image data acquired through the image acquisition unit (1310) can be processed by the image processing unit (1320).

The image processing unit (1320) can perform processing on image data received from the image acquisition unit (1310). The image processing unit (1320) can perform various image processing operations, such as decoding, scaling, noise filtering, frame rate conversion, or resolution conversion on the image data.

The display (1330) can display image data processed by the image processing unit (1320). At least a portion of the display (1330) may be combined with at least one of the front area, the side area, and the rear area of the electronic device (100) in the form of a flexible display. The flexible display may be characterized by being able to be bent, folded, or rolled without damage through a thin and flexible substrate like paper.

The display (1330) can be implemented as a layered touch screen in combination with a touch panel (1381). The touch screen can have not only a display function but also a function to detect a touch input location, a touched area, and even a touch input pressure, and can also have a function to detect not only a real touch but also a proximity touch.

The communication unit (1340) can communicate with various types of external devices according to various types of communication methods. The communication unit (1340) can include at least one of a Wi-Fi chip (1341), a Bluetooth chip (1342), a wireless communication chip (1343), and an NFC chip (1344). The processor (1390) can communicate with an external server or various external devices using the communication unit (1340).

The audio processing unit (1360) is a component that performs processing on audio data. The audio processing unit (1360) can perform various processing operations on audio data, such as decoding, amplification, and noise filtering. Audio data processed by the audio processing unit (1360) can be output to the audio output unit (1370).

The audio output unit (1370) is configured to output various audio data that have undergone various processing operations, such as decoding, amplification, and noise filtering, by the audio processing unit (1360), as well as various notification sounds or voice messages. In particular, the audio output unit (1370) may be implemented as a speaker, but this is only one embodiment, and may be implemented as an output terminal capable of outputting audio data.

The input unit (1380) can receive various user inputs and transmit them to the processor (1390). The user input unit (1380) can include, for example, a touch panel (1381), a (digital) pen sensor (1382), or keys (1383). The touch panel (1381) can use, for example, at least one of a capacitive method, a pressure-sensitive method, an infrared method, or an ultrasonic method. In addition, the touch panel (1381) may further include a control circuit. The touch panel (1381) can further include a tactile layer to provide a tactile response to the user. The (digital) pen sensor (1382) can be, for example, part of the touch panel or include a separate recognition sheet. The keys (1383) can include, for example, a physical button, an optical key, or a keypad.

The processor (1390) (or control unit) can control the overall operation of the electronic device (100) using various programs stored in the memory (1350).

The processor (1390) may be composed of a RAM (1391), a ROM (1392), a graphics processing unit (1393), a main CPU (1394), first to n interfaces (1395-1 to 1395-n), and a bus (1396). At this time, the RAM (1391), the ROM (1392), the graphics processing unit (1393), the main CPU (1394), the first to n interfaces (1395-1 to 1395-n), etc. may be connected to each other via the bus (1396).

Meanwhile, the function related to artificial intelligence according to the present disclosure is operated through a processor (120) and a memory (110). The processor (120) may be composed of one or more processors. In this case, one or more processors may be a general-purpose processor such as a CPU, an AP, a DSP (Digital Signal Processor), a graphics-only processor such as a GPU, a VPU (Vision Processing Unit), or an artificial intelligence-only processor such as an NPU. One or more processors control to process input data according to predefined operation rules or artificial intelligence models stored in the memory. Alternatively, when one or more processors are artificial intelligence-only processors, the artificial intelligence-only processor may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

The predefined operation rules or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that the basic artificial intelligence model is trained using a learning algorithm using a plurality of learning data, thereby creating a predefined operation rules or artificial intelligence model set to perform a desired characteristic (or purpose). This learning may be performed on the device itself on which the artificial intelligence according to the present disclosure is performed, or may be performed through a separate server and/or system. Examples of the learning algorithm include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values, and performs neural network operations through operations between the operation results of the previous layer and the multiple weights. The multiple weights of the multiple neural network layers may be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights may be updated so that the loss value or cost value obtained from the artificial intelligence model is reduced or minimized during the learning process. The artificial neural network may include a deep neural network (DNN), and examples thereof include, but are not limited to, a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), or deep Q-networks.

FIG. 14 is a flowchart illustrating the operation of an electronic device according to an embodiment of the present disclosure.

The electronic device can receive a user command (S1610). Here, the user command can be text entered via a keyboard or on-screen keyboard, or a picture or text entered via a stylus or the like. Additionally, the user command can also be the user's voice entered via a microphone.

In addition, the electronic device can classify user commands (S1620). Specifically, the electronic device can classify categories of user commands using the first artificial intelligence model.

To this end, when a user command is received, the electronic device (100) can first identify a text corresponding to the user command from among a plurality of pre-stored texts using a first artificial intelligence model. Furthermore, the electronic device (100) can classify the category of the user command based on the category information matched to each text.

Thereafter, the electronic device can provide response information to a user command based on the classification result (S1630).

Specifically, the electronic device can provide information corresponding to the user command as response information from among a plurality of pieces of information included in the category of the user command using the first artificial intelligence model.

For example, if a text such as "taxi search" is stored in an electronic device and "find a taxi" is received, the electronic device can identify "taxi search" as the text corresponding to "find a taxi" among the plurality of stored texts. In addition, the electronic device can classify the "transportation" category matching "taxi search" as a category of the user command, and provide information related to "taxi search" among the plurality of pieces of information included in the "transportation" category as response information. For example, the electronic device can provide information related to available taxis around the user as response information.

Meanwhile, if the category of the user command is classified as an error category, the electronic device may provide error information indicating that the information requested by the user command cannot be found. In other words, the electronic device may reject the user command.

Accordingly, unlike conventional electronic devices, the electronic device according to one embodiment of the present disclosure can prevent cases where response information is provided that is completely unrelated to the user's intention.

Meanwhile, the methods according to the various embodiments of the present invention described above can be implemented only with software/hardware upgrades for existing electronic devices.

In addition, the various embodiments of the present invention described above can also be performed through an embedded server provided in an electronic device or an external server.

The control methods for electronic devices according to the various embodiments described above may be implemented as programs and stored on various recording media. That is, a computer program capable of executing the various control methods described above by being processed by various processors may be stored on a recording medium and used.

A non-transitory readable medium refers to a medium that permanently stores data and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, the various applications or programs described above may be stored and provided on non-transitory readable media, such as a CD, DVD, hard disk, Blu-ray disc, USB, memory card, or ROM.

Although the preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and various modifications may be made by those skilled in the art without departing from the gist of the present invention as claimed in the claims. Furthermore, such modifications should not be understood individually from the technical idea or prospect of the present invention.

100: Electronic devices
110: Memory
120: Processor

Claims (16)

In electronic devices,
Memory that stores computer executable instructions; and
A processor that determines a text corresponding to a received user command by executing the computer-executable instructions, and if the text is classified as corresponding to one of a plurality of pre-stored texts through a first artificial intelligence model, provides response information for the user command, and if the text is classified as an error through the first artificial intelligence model, provides error information for the user command;
The above first artificial intelligence model is,
If the text corresponding to the user command is a similar text that has a different entity and intent from at least one of the plurality of pre-stored texts, the text is classified as the error.
The above similar text is generated by the second artificial intelligence model,
The above first artificial intelligence model is,
Determine whether the similar text input to the first artificial intelligence model corresponds to one of the plurality of pre-stored texts, and determine whether to learn based on the determination result.
The above second artificial intelligence model is,
An electronic device that learns to generate a new similar text having a higher degree of similarity than the degree to which the similar text is similar to at least one text, when the first artificial intelligence model determines that the similar text does not correspond to the plurality of pre-stored texts. In the first paragraph,
The above first artificial intelligence model is,
An electronic device, wherein a model is obtained by training a model that determines a text corresponding to the user command based on the plurality of pre-stored texts, and classifies the similar text as an error. delete delete In the first paragraph,
The above first artificial intelligence model is,
An electronic device that learns to classify the similar text as an error if it is determined that the similar text corresponds to one of the plurality of pre-stored texts, and does not perform the learning if it is determined that the similar text does not correspond to one of the plurality of pre-stored texts. delete In the first paragraph,
The above first artificial intelligence model is,
An electronic device that receives the new similar text generated by the second artificial intelligence model, determines whether the new similar text corresponds to one of the plurality of pre-stored texts, and determines whether to learn based on the determination result. In the first paragraph,
The first and second artificial intelligence models are electronic devices that are Generative Adversarial Networks (GANs). In a method for controlling an electronic device,
a step of determining a text corresponding to a received user command; and
A step of providing response information for the user command when the text is classified as corresponding to one of a plurality of pre-stored texts through the first artificial intelligence model, and providing error information for the user command when the text is classified as an error through the first artificial intelligence model;
The above first artificial intelligence model is,
If the text corresponding to the user command is a similar text that has a different entity and intent from at least one of the plurality of pre-stored texts, the text is classified as the error.
The above similar text is generated by the second artificial intelligence model,
The above first artificial intelligence model is,
Determine whether the similar text input to the first artificial intelligence model corresponds to one of the plurality of pre-stored texts, and determine whether to learn based on the determination result.
The above second artificial intelligence model is,
A control method of an electronic device, wherein, if the first artificial intelligence model determines that the similar text does not correspond to the plurality of pre-stored texts, the first artificial intelligence model learns to generate a new similar text having a higher degree of similarity than the degree of similarity of the similar text to the at least one text. In paragraph 9,
The above first artificial intelligence model is,
A control method for an electronic device, wherein the model is obtained by training a model that determines a text corresponding to the user command based on the plurality of pre-stored texts, and classifies the similar text as an error. delete delete In paragraph 9,
The above first artificial intelligence model is,
A control method of an electronic device, wherein learning is performed to classify the similar text as an error if the similar text is determined to correspond to one of the plurality of pre-stored texts, and the learning is not performed if the similar text is determined not to correspond to the plurality of pre-stored texts. delete In paragraph 9,
The above first artificial intelligence model is,
A control method of an electronic device, which receives the new similar text generated by the second artificial intelligence model, determines whether the new similar text corresponds to one of the plurality of pre-stored texts, and determines whether to learn based on the determination result. In paragraph 9,
A method for controlling an electronic device, wherein the first and second artificial intelligence models are GAN (Generative Adversarial Network).