Question generation by intent prediction

1. A computer system for question generation, the computer system comprising:

one or more computer processors;

one or more computer readable storage devices; and

stored program instructions on the one or more computer readable storage devices for execution by the one or more computer processors, the stored program instructions comprising:

program instructions to receive user utterance data;

program instructions to determine an intent confidence vector for the user utterance data;

program instructions to train the next user-intent prediction model comprising:

program instructions to receive historic chatbot conversation data, wherein the chatbot conversation data comprises labeled user-intents and intent confidence vectors for each user utterance of each conversation;

program instructions to separate the chatbot conversation data into training data and validation data;

program instructions to generate a matrix including all the intent confidence vectors for all historic conversation training data;

program instructions to train a machine learning model to classify user-intents using the matrix of training data;

program instructions to drop out at least one intent confidence vector during training;

program instructions to validate the machine learning model using the validation data; and

program instructions to provide a trained next intent predictor mode for use in generating questions:

program instructions to predict, by the trained next user-intent prediction model, a next user-intent confidence vector using the intent confidence vector;

program instructions to generate a next question set using the next user-intent confidence vector;

program instructions to adjust a confidence value of the next user-intent confidence vector according to value ranking for a predefined type of a next user intent provided by a productive value ranking algorithm;

program instructions to generate an ordered version of the next question set ordered according to next user intent confidence vector values adjusted according to the productive value ranking algorithm; and

program instructions to output a subset of the ordered version having a highest value to the user.

2. The computer system according to claim 1 , the stored program instructions further comprising:

program instructions to predict a next user-intent confidence vector using a sequence of intent confidence vectors determined from a sequence of received user utterance data.

3. The computer system according to claim 1 , the stored program instructions further comprising program instructions to train a long short term memory (LSTM) neural network with L1 regularization, as the next user-intent prediction model, and program instructions to provide the trained LSTM neural network or use predicting a next user-intent.

4. The computer system according to claim 3 , the stored program instructions further comprising program instructions to train the LSTM neural network using a least absolute shrinkage and selection operator for regularization.

5. A computer program product for question generation, the computer program product comprising one or more computer readable storage devices and collectively stored program instructions on the one or more computer readable storage devices, the stored program instructions comprising:

program instructions to receive user utterance data;

program instructions to determine an intent confidence vector for the user utterance data;

program instructions to train the next user-intent prediction model comprising:

program instructions to receive historic chatbot conversation data, wherein the chatbot conversation data comprises labeled user-intents and intent confidence vectors for each user utterance of each conversation;

program instructions to separate the chatbot conversation data into training data and validation data;

program instructions to generate a matrix including all the intent confidence vectors for all historic conversation training data;

program instructions to train a machine learning model to classify user-intents using the matrix of training data;

program instructions to drop out at least one intent confidence vector during training;

program instructions to validate the machine learning model using the validation data; and

program instructions to provide a trained next intent predictor mode for use in generating questions:

program instructions to predict, by the trained next user-intent prediction model, a next user-intent confidence vector using the intent confidence vector;

program instructions to generate a next question set using the next user-intent confidence vector;

program instructions to adjust a confidence value of the next user-intent confidence vector according to value ranking for a predefined type of a next user intent provided by a productive value ranking algorithm;

program instructions to generate an ordered version of the next question set ordered according to next user intent confidence vector values adjusted according to the productive value ranking algorithm; and

program instructions to output a subset of the ordered version having a highest value to the user.

6. The computer program product according to claim 5 , the stored program instructions further comprising:

program instructions to predict a next user-intent confidence vector using a sequence of intent confidence vectors determined from a sequence of received user utterance data.

7. The computer program product according to claim 5 , the stored program instructions further comprising program instructions to train a long short term memory (LSTM) neural network with L1 regularization, as the next user-intent prediction model, and program instructions to provide the trained LSTM neural network or use predicting a next user-intent.

8. The computer program product according to claim 7 , the stored program instructions further comprising program instructions to train the LSTM neural network using a least absolute shrinkage and selection operator for regularization.

9. The computer program product according to claim 5 , wherein

program instructions to receive user utterance data comprises receiving data from a plurality of conversations.

10. A computer implemented method for generating questions, the method comprising:

receiving, by one or more computer processors, user utterance data;

determining, by the one or more computer processors, an intent confidence vector for the user utterance data;

training, by the one or more computer processors, the next user-intent prediction model by:

receiving, by the one or more computer processors, historic chatbot conversation data, wherein the chatbot conversation data comprises labeled user-intents and intent confidence vectors for each user utterance of each conversation;

separating, by the one or more computer processors, the chatbot conversation data into training data and validation data;

generating, by the one or more computer processors, a matrix including all the intent confidence vectors for all historic conversation training data;

training, by the one or more computer processors, a machine learning model to classify user-intents using the matrix of training data;

dropping out, by the one or more computer processors, at least one intent confidence vector during training:

validating, by the one or more computer processors, the trained machine learning model using the validation data; and

providing, by the one or more computer processors, a trained next intent predictor mode for use in generating questions:

predicting, by the one or more computer processors through the trained next user-intent prediction model, a next user-intent confidence vector using the intent confidence vector;

generating, by the one or more computer processors, a next question set using the next user-intent confidence vector;

adjusting, by the one or more processors, a confidence value of the next user-intent confidence vector according to value ranking for a predefined type of a next user intent provided by a productive value ranking algorithm;

generating, by the one or more processors, an ordered version of the next question set ordered according to next user intent confidence vector values adjusted according to the productive value ranking algorithm; and

outputting, by the one or more processors, a subset of the ordered version having a highest value to the user.

11. The computer implemented method according to claim 10 , further comprising:

predicting, by the one or more computer processors, a next user-intent confidence vector using a sequence of intent confidence vectors determined from a sequence of received user utterance data.

12. The computer implemented method according to claim 10 , further comprising training, by the one or more computer processors, a long short term memory (LSTM) neural network with L1 regularization, as the next user-intent prediction model, and providing the trained LSTM neural network or use predicting a next user-intent.

13. The computer implemented method of claim 12 , further comprising training, by the one or more computer processors, the LSTM neural network using a least absolute shrinkage and selection operator for regularization.

14. The computer implemented method according to claim 10 , wherein

receiving user utterance data comprises receiving data from a plurality of conversations.