Electronic apparatus and method for controlling thereof
An electronic apparatus is provided. The electronic apparatus includes a communication interface, a memory, and a processor, wherein the processor is configured to acquire a user command from the user and control the communication interface to transmit the user command to a plurality of external devices, receive information on a first question generated based on the user command and information on first response to the generated first question acquired from users of each of the plurality of external devices, identify whether there a conflict between each first response occurs, acquire information on a subject to be re-questioned, control the communication interface to transmit information on the conflict to at least one external device identified, receive information on a second response to a second question generated, and acquire a final response based on the information on the first response and the information on the second response and output the final response.
1. A method of controlling an electronic apparatus comprising:
acquiring a user command from a user and transmitting the user command to a plurality of external devices;
receiving, from each external device of the plurality of external devices, information on a first question generated by the external device based on the user command and information on a first response acquired by the external device in response to the generated first question;
analyzing each of the received information on the first response;
identifying, based on the analyzing, whether a conflict between each of the first response occurs;
based on the identifying that the conflict occurs, acquiring information on a subject to be re-questioned based on the received information on the generated first question and the received information on the first response;
identifying at least one external device of the plurality of external devices, based on the information on the subject to be re-questioned;
transmitting information on the conflict to the identified at least one external device;
receiving, from the identified at least one external device, information on a second response acquired by the identified at least one external device in response to a second question generated based on the information on the conflict;
determining a final response based on the information on the first response and the information on the second response; and
outputting the final response,
wherein the conflict is that information acquired from each of the plurality of external devices conflicts with each other,
wherein the information includes a graph indicating a degree of conflict between pieces of information,
wherein the information on the generated first question and the information on the first response includes specific information and remaining information,
wherein a specific value is added to each of the specific information and the remaining information,
wherein the identifying of whether the conflict between each of the first response occurs comprises:
inputting a predetermined token and the information on the first response into a first neural network model and acquiring an embedding vector corresponding to the predetermined token, and
based on the embedding vector being a predetermined value, identifying that the conflict between each of the first response occurs, and
wherein the acquiring of the information on the subject to be re-questioned comprises:
acquiring an attention score indicating a degree of the conflict between each of the first response based on a weight value of the first neural network mode,
acquiring a vector corresponding to each of the plurality of external devices by inputting the predetermined token and the information on the first response into the first neural network model,
acquiring conflict information on the conflict between the first response based on the attention score and the vector, and
acquiring the information on the subject to be asked by inputting the conflict information into a second neural network model.
2. The method of claim 1 , wherein the first neural network model adapts to learn to identify whether a plurality of input text conflict with each other.
3. The method of claim 1 , wherein the acquiring of the attention score comprises acquiring a weight value for calculating data included in a last layer of the first neural network model as the attention score.
4. The method of claim 1 , wherein the second neural network model is configured to be a graph neural network model learned to minimize a conversation cost between the plurality of external devices and users of each of the plurality of external devices.
5. The method of claim 1 , further comprising:
identifying whether a conflict between the second responses occurs by inputting the predetermined token and the information on the second response into the first neural network model,
wherein the acquiring of the final response comprises, based on the conflict between the second responses being not occurred, acquiring the final response based on the information on the first response and the information on the second response.
6. The method of claim 1 , wherein the acquiring of the user command and transmitting the user command to the plurality of external devices comprises:
dividing the user command into a plurality of commands;
identifying the plurality of external devices corresponding to each of the plurality of divided commands; and
transmitting the divided plurality of commands to the identified plurality of external devices.
7. The method of claim 6 , wherein the identifying of the plurality of external devices comprises:
identifying keywords included in the divided plurality of commands; and
identifying the plurality of external devices corresponding to the identified keywords based on a lookup table in which the keywords are matched with external devices previously stored in the electronic apparatus.
8. An electronic apparatus comprising:
a communication interface;
memory storing at least one instruction; and
a processor,
wherein the processor, when executing the at least one instruction, is configured to:
acquire a user command from a user and control the communication interface to transmit the user command to a plurality of external devices,
receive, from each external device of the plurality of external devices, information on a first question generated by the external device based on the user command and information on a first response acquired by the external device in response to the generated first question,
analyze each of the received information on the first response,
identify, based on the analysis, whether there a conflict between each of the first response occurs,
based on identification that the conflict occurs, acquire information on a subject to be re-questioned based on the received information on the generated first question and the received information on the first response,
identify at least one external device of the plurality of external devices, based on the information on the subject to be re-questioned,
control the communication interface to transmit information on the conflict to the identified at least one external device,
receive, from the identified at least one external device, information on a second response acquired by the identified at least one external device in response to a second question generated based on the information on the conflict,
determine a final response based on the information on the first response and the information on the second response, and
output the final response,
wherein the conflict is that information acquired from each of the plurality of external devices conflicts with each other,
wherein the information includes a graph indicating a degree of conflict between pieces of information,
wherein the information on the generated first question and the information on the first response includes specific information and remaining information,
wherein a specific value is added to each of the specific information and the remaining information,
wherein the processor, when executing the at least one instruction, is further configured to:
input a predetermined token and the information on the first response into a first neural network model and acquiring an embedding vector corresponding to the predetermined token, and
based on the embedding vector is a predetermined value, identify the conflict between each of the first response occurs, and
wherein the processor, when executing the at least one instruction, is further configured to:
acquire an attention score indicating a degree of the conflict between each of the first response based on a weight value of the first neural network model,
acquire a vector corresponding to each of the plurality of external devices by inputting the predetermined token and the information on the first response into the first neural network model,
acquire conflict information on the conflict between each of the first response based on the attention score and the vector, and
acquire the information on the subject to be asked by inputting the conflict information into a second neural network model.
9. The electronic apparatus of claim 8 , wherein the first neural network model adapts to learn to identify whether a plurality of input text conflict with each other.
10. The electronic apparatus of claim 8 , wherein the processor, when executing the at least one instruction, is further configured to acquire a weight value for calculating data included in a last layer of the first neural network model as the attention score.
11. The electronic apparatus of claim 8 , wherein the second neural network model is configured to be a graph neural network model learned to minimize a conversation cost between the plurality of external devices and users of each of the plurality of external devices.
12. The electronic apparatus of claim 8 , wherein the first neural network model comprises a plurality of neural network layers.
13. The electronic apparatus of claim 12 ,
wherein each of the plurality of neural network layers has a plurality of weight values, and
wherein the processor, when executing the at least one instruction, is further configured to perform a neural network operation through an operation between an operation result of a previous layer and the plurality of weight values.
14. The electronic apparatus of claim 8 , wherein the attention score indicating the degree of the conflict between each of the first responses further indicates a degree to which each piece information of each of the first response conflicts based on an intermediate result value.