Computer-implemented method, system and computer program product for providing an application interface
A computer-implemented method is provided for providing an interface between a frontend application configured to receive one or more user inputs in a natural language and a backend system configured to provide a service to a user. The method may comprise: receiving a text input in the natural language via the frontend application; performing computation using the received text input and a neural network, the neural network comprising: a convolutional layer configured to generate feature values by applying one or more filters to an input matrix obtained from the received text input, each of the one or more filters having a window size corresponding to one or more words contained in the text input; one or more long short-term memory, LSTM, layers configured to generate output values by processing at least a part of the feature values generated by the convolutional layer; and an output layer configured to generate an output vector based on the output values generated by the one or more LSTM layers, the output vector representing an action to be performed by the backend system in response to the received text input; identifying the action represented by the output vector generated as a result of the computation; communicating the identified action to the backend system for the backend system to perform the identified action; and providing the frontend application with a text output in the natural language based on the identified action.
1. A computer-implemented method for providing an interface between a frontend application configured to receive one or more user inputs in a natural language and a backend system configured to provide a service to a user, the method comprising:
receiving a text input in the natural language via the frontend application;
performing computation using the received text input and a neural network, the neural network comprising:
a convolutional layer configured to generate feature values by applying one or more filters to an input matrix obtained from the received text input, each of the one or more filters having a window size corresponding to one or more words contained in the text input;
one or more long short-term memory, LSTM, layers configured to:
receive a set of additional input parameters from a source external to the neural network, the additional input parameters relating to an action to be provided by the backend system, and
generate output values by processing at least a part of the feature values generated by the convolutional layer and the additional data,
wherein the set of additional input parameters is further used to train the neural network; and
an output layer configured to generate an output vector based on the output values generated by the one or more LSTM layers, the output vector representing the action to be performed by the backend system in response to the received text input;
identifying the action represented by the output vector generated as a result of the computation;
communicating the identified action to the backend system for the backend system to perform the identified action; and
providing the frontend application with a text output in the natural language based on the identified action.
2. The computer-implemented method according to claim 1 , wherein the convolutional layer comprised in the neural network is configured to apply a plurality of filters to the input matrix for generating the feature values, the plurality of filters having different window sizes.
3. The computer-implemented method according to claim 2 , wherein the one or more LSTM layers comprised in the neural network are configured to process all the feature values generated by the convolutional layer for generating the output values.
4. The computer-implemented method according to claim 1 , wherein the neural network ( 50 a ) further comprises a max-pooling layer configured to perform a max-pooling operation to the feature values generated by the convolutional layer; and
wherein the one or more LSTM layers comprised in the neural network are configured to process values selected from the feature values in the max-pooling operation for generating the output values.
5. The computer-implemented method according to claim 4 , wherein the one or more LSTM layers comprised in the neural network are configured to process not only at least the part of the feature values generated by the convolutional layer but also additional input parameters relating to the service provided by the backend system for generating the output values.
6. The computer-implemented method according to claim 5 , further comprising:
receiving a sequence of possible text inputs and information indicating an action to be taken by the backend system in response to each of the possible text inputs; and
training the neural network using a sequence of input matrices obtained from the sequence of the possible text inputs and the information indicating the action to be taken by the backend system in response to each of the possible text inputs.
7. A system for providing an interface between a frontend application configured to receive one or more user inputs in a natural language and a backend system configured to provide a service to a user, the system comprising one or more processors configured to:
receive a text input in the natural language via the frontend application;
perform computation using the received text input and a neural network, the neural network comprising:
a convolutional layer configured to generate feature values by applying one or more filters to an input matrix obtained from the received text input, each of the one or more filters having a window size corresponding to one or more words contained in the text input;
one or more long short-term memory, LSTM, layers configured to:
receive a set of additional input parameters from a source external to the neural network, the additional input parameters relating to an action to be provided by the backend system, and
generate output values by processing at least a part of the feature values generated by the convolutional layer and the additional data,
wherein the set of additional input parameters is further used to train the neural network; and
an output layer configured to generate an output vector based on the output values generated by the one or more LSTM layers, the output vector representing an action to be performed by the backend system in response to the received text input;
identify the action corresponding to the output vector generated as a result of the computation;
communicate the identified action to the backend system for the backend system to perform the identified action; and
provide the frontend application with a text output in the natural language based on the identified action.
8. The system according to claim 7 , wherein the convolutional layer comprised in the neural network is configured to apply a plurality of filters to the input matrix for generating the feature values, the plurality of filters having different window sizes; and
wherein the one or more LSTM layers comprised in the neural network may be configured to process all the feature values generated by the convolutional layer for generating the output values.
9. The system according to claim 8 , wherein the one or more LSTM layers comprised in the neural network are configured to process not only at least the part of the feature values generated by the convolutional layer but also additional input parameters relating to the service provided by the backend system for generating the output values.
10. The system according to claim 8 , wherein the one or more processers are further configured to:
receive a sequence of possible text inputs and information indicating an action to be taken by the backend system in response to each of the possible text inputs; and
train the neural network using a sequence of input matrices obtained from the sequence of the possible text inputs and the information indicating the action to be taken by the backend system in response to each of the possible text inputs.
11. A system for training a neural network to provide an interface between a frontend application configured to receive one or more user inputs in a natural language and a backend system configured to provide a service to a user, the system comprising one or more processors configured to:
receive a sequence of possible text inputs and information indicating an action to be taken by the backend system in response to each of the possible text inputs; and
train the neural network using a sequence of input matrices obtained from the sequence of the possible text inputs and the information indicating the action to be taken by the backend system in response to each of the possible text inputs,
wherein the neural network comprises:
a convolutional layer configured to generate feature values by applying one or more filters to an input matrix obtained from a text input, each of the one or more filters having a window size corresponding to one or more words contained in the text input;
one or more long short-term memory, LSTM, layers configured to:
receive a set of additional input parameters from a source external to the neural network, the additional input parameters relating to an action to be provided by the backend system, and
generate output values by processing at least a part of the feature values generated by the convolutional layer and the additional data,
wherein the set of additional input parameters is further used to train the neural network; and
an output layer configured to generate an output vector based on the output values generated by the one or more LSTM layers, the output vector representing an action to be performed by the backend system in response to the text input.