Electronic device and computer readable storage medium for control recommendation
An electronic device is provided. The electronic device includes an interface, a memory, and a processor and configured to provide a control recommendation of an external electronic device using a learning model. The learning model is configured to generate a first output vector by encoding sequential control information about a user using a transformer and summarizing the encoded sequential control information using a query vector, and output a second output vector by encoding the first output vector using a transformer and summarizing the encoded first output vector using time information.
1 . An electronic device comprising:
an interface;
a memory configured to store a learning model; and
a processor configured to provide a control recommendation for an external electronic device by using the learning model stored in the memory,
wherein the learning model comprises:
an input layer that generates a plurality of first embedding vectors corresponding to an input sequence including a series of control histories of a user on a plurality of external electronic devices by applying embedding weights to the input sequence,
a first encoding layer that outputs a plurality of first output vectors by using one or more transformers to generate a plurality of respective first encoded vectors from the plurality of first embedding vectors, applying first weights to the plurality of first encoded vectors, and adding the plurality of first encoded vectors to which the first weights have been applied, and
a second encoding layer that outputs a second output vector by adding position information to the plurality of first output vectors to generate a plurality of second embedding vectors, using the one or more transformers to generate a plurality of second encoded vectors from the plurality of second embedding vectors, applying second weights to values of the plurality of second encoded vectors, and adding the plurality of second encoded vectors to which the second weights have been applied,
wherein the first weights are based on a query vector and first trained parameters,
wherein the second weights are based on time information and second trained parameters, and
wherein the first trained parameters and the second trained parameters are learned such that a loss between training data for the learning model and the control recommendation based on the learning model is minimized.
2 . The electronic device of claim 1 , wherein each of the control histories includes:
information about a target external electronic device for a control;
information on a control function for the target external electronic device; and
information on a control time of the target external electronic device.
3 . The electronic device of claim 2 , wherein the information on the control time includes information on a control day and a control hour.
4 . The electronic device of claim 1 ,
wherein the embedding weights are normalized through transfer learning using a plurality of pieces of routine data set by a plurality of users, and
wherein each of the plurality of pieces of routine data includes a control sequence of a plurality of external electronic devices set by one user.
5 . The electronic device of claim 1 , wherein the processor is further configured to, in response to sensing a trigger event, provide the control recommendation to the user using the learning model.
6 . The electronic device of claim 5 , wherein the processor is further configured to:
acquire an utterance of the user through the interface; and
sense the trigger event if the utterance of the user includes an intent corresponding to control of an external electronic device.
7 . The electronic device of claim 5 , wherein the processor is further configured to sense the trigger event if a call of a voice agent of the user is sensed.
8 . The electronic device of claim 1 , wherein the time information corresponds to a time to provide the control recommendation.
9 . The electronic device of claim 1 , wherein the second output vector includes information on control probabilities of a plurality of external electronic devices at a time to provide the control recommendation.
10 . The electronic device of claim 9 , wherein the processor is further configured to identify a target device related to the control recommendation based on the control probabilities.
11 . A non-transitory computer readable storage medium storing instructions and a learning model, the instructions, when executed by a processor of an electronic device, causing the electronic device to provide a control recommendation for an external electronic device using the learning model, the learning model comprising:
an input layer that generates a plurality of first embedding vectors corresponding to an input sequence including a series of control histories of a user on a plurality of external electronic devices by applying embedding weights to the input sequence;
a first encoding layer that outputs a plurality of first output vectors by using one or more transformers to generate a plurality of respective first encoded vectors from the plurality of first embedding vectors, applying first weights to the plurality of first encoded vectors, and adding the plurality of first encoded vectors to which the first weights have been applied; and
a second encoding layer that outputs a second output vector by adding position information to the plurality of first output vectors to generate a plurality of second embedding vectors, using the one or more transformers to generate a plurality of second encoded vectors from the plurality of second embedding vectors, applying second weights to values of the plurality of second encoded vectors, and adding the plurality of second encoded vectors to which the second weights have been applied,
wherein the first weights are based on a query vector and first trained parameters,
wherein the second weights are based on time information and second trained parameters, and
wherein the first trained parameters and the second trained parameters are learned such that a loss between training data for the learning model and the control recommendation based on the learning model is minimized.
12 . The non-transitory computer readable storage medium of claim 11 , wherein each of the control histories includes:
information about a target external electronic device to be controlled;
information on a control function for the target external electronic device; and
information on a control time of the target external electronic device.
13 . The non-transitory computer readable storage medium of claim 12 , wherein the information on the control time includes information on a control day and a control hour.
14 . The non-transitory computer readable storage medium of claim 11 ,
wherein the embedding weights are normalized through transfer learning using a plurality of pieces of routine data set by a plurality of users, and
wherein each of the plurality of pieces of routine data includes a control sequence of a plurality of external electronic devices set by one user.
15 . The non-transitory computer readable storage medium of claim 11 , wherein the instructions, when executed by the processor, further cause the electronic device to, in response to sensing a trigger event, provide the control recommendation to the user using the learning model.
16 . The non-transitory computer readable storage medium of claim 15 , wherein the instructions, when executed by the processor, further cause the electronic device to:
acquire an utterance of the user through an interface of the electronic device; and
sense the trigger event if the utterance of the user includes an intent corresponding to control of an external electronic device.
17 . The non-transitory computer readable storage medium of claim 15 , wherein the instructions, when executed by the processor, further cause the electronic device to sense the trigger event if a call of a voice agent of the user is sensed.
18 . The non-transitory computer readable storage medium of claim 15 , wherein the time information corresponds to a time to provide the control recommendation.
19 . The non-transitory computer readable storage medium of claim 11 , wherein the second output vector includes information on control probabilities of a plurality of external electronic devices at a time to provide the control recommendation.
20 . The non-transitory computer readable storage medium of claim 19 , wherein the instructions, when executed by the processor, further cause the electronic device to identify a target device related to the control recommendation based on the control probabilities.