Generating workflow representations using reinforced feedback analysis
Generating visual workflow representations by receiving data including text instructions, identifying actions in the instructions, generating a mapping of the actions according to a generative model, the mapping including an action sequence, providing the mapping to a user, receiving feedback from the user, altering the generative model according to the feedback, and generating a revised mapping according to the feedback.
1 . A method for generating visual workflow representations, the method comprising:
receiving, by one or more computer processors, data including workflow documentation comprising text and audio data;
converting the audio data to text;
identifying, by the one or more computer processors, entities and actions using a topic classification process of natural language understanding (NLU) and part-of-speech tagging performed using a Hidden Markov Model with Viterbi decoding, and forming dictionary pairings of entity-action terms;
determining, by the one or more computer processors, relationships between the identified entities and actions using the NLU output, including associating each action with a corresponding command-line interface (CLI) command extracted from the workflow documentation;
generating, by the one or more computer processors, a graph from the relationships, the graph comprising nodes representing the entities and directed edges representing the actions, each directed edge being annotated with its associated CLI command;
training, by the one or more computer processors, a generative adversarial network (GAN) comprising a generator and a discriminator via gradient descent and back-propagation using labeled data from a target domain:
generating, by the GAN, graphical workflow elements corresponding to the graph;
arranging the graphical workflow elements by applying a Hamiltonian Monte Carlo (HMC) method to the graphical workflow elements and selecting a sample having a highest probability as an initial layout;
generating and providing, by the one or more computer processors, a mapping to a user, the mapping comprising the arranged graphical workflow elements with executable CLI links that, when selected by the user, invoke the associated CLI commands to implement workflow tasks directly from the mapping;
receiving, by the one or more computer processors, user feedback comprising corrections to entity-action pairings and layout attributes including adjacency, arrow type, and line type feedback from the user;
computing, by the one or more computer processors, a reinforcement-learning reward from the user feedback and altering generator and discriminator node weightings via back-propagation according to the reward; and
regenerating, by the one or more computer processors, the graphical workflow elements and re-arranging the elements via the HMC sampling procedure to generate a revised mapping according to the user feedback.
2 . The method according to claim 1 , further comprising analyzing, by the one or more computer processors, the feedback according to a dimension selected from the group consisting of tone, sentiment, confusion, and combinations thereof.
3 . A computer program product for generating visual workflow representations, 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 data including workflow documentation comprising text and audio data;
converting the audio data to text
identify entities and actions using a topic classification process of natural language processing (NLU) and part-of-speech tagging performed using a Hidden Markov Model with Viterbi decoding, and forming dictionary pairings of entity-action terms;
determine relationships between the identified entities and actions using the NLU output, including associating each action with a corresponding command-line interface (CLI) command extracted from the workflow documentation entities and actions;
generate a graph from the relationships, the graph comprising nodes representing the entities and directed edges representing the actions, each directed edge being annotated with its associated CLI command;
train a generative adversarial network (GAN) comprising a generator and a discriminator via gradient descent and back-propagation using labeled data from a target domain:
generate graphical workflow elements corresponding to the graph;
arrange the graphical workflow elements by applying a Hamiltonian Monte Carlo (HMC) method to the graphical workflow elements and selecting a sample having a highest probability as an initial layout;
generate and provide a mapping to a user, the mapping comprising the arranged graphical workflow elements with executable CLI links that, when selected by the user, invoke the associated CLI commands to implement workflow tasks directly from the mapping;
receive user feedback comprising corrections to entity-action pairings and layout attributes including adjacency, arrow type, and line type;
computing, by the one or more computer processors, a reinforcement-learning reward from the user feedback and altering generator and discriminator node weightings via back-propagation according to the reward; and
regenerate the graphical workflow elements and re-arranging the elements via the HMC sampling procedure to generate a revised mapping according to the user feedback.
4 . The computer program product according to claim 3 , the stored program instructions further comprising program instructions to analyze the feedback according to a dimension selected from the group consisting of tone, sentiment, confusion, and combinations thereof.
5 . A computer system for generating visual workflow representations, 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 data including workflow documentation comprising text and audio data;
converting the audio data to text
identify entities and actions using a topic classification process of natural language processing (NLU) and part-of-speech tagging performed using a Hidden Markov Model with Viterbi decoding, and forming dictionary pairings of entity-action terms;
determine relationships between the identified entities and actions using the NLU output, including associating each action with a corresponding command-line interface (CLI) command extracted from the workflow documentation entities and actions;
generate a graph from the relationships, the graph comprising nodes representing the entities and directed edges representing the actions, each directed edge being annotated with its associated CLI command;
train a generative adversarial network (GAN) comprising a generator and a discriminator via gradient descent and back-propagation using labeled data from a target domain:
generate graphical workflow elements corresponding to the graph;
arrange the graphical workflow elements by applying a Hamiltonian Monte Carlo (HMC) method to the graphical workflow elements and selecting a sample having a highest probability as an initial layout;
generate and provide a mapping to a user, the mapping comprising the arranged graphical workflow elements with executable CLI links that, when selected by the user, invoke the associated CLI commands to implement workflow tasks directly from the mapping;
receive user feedback comprising corrections to entity-action pairings and layout attributes including adjacency, arrow type, and line type;
computing, by the one or more computer processors, a reinforcement-learning reward from the user feedback and altering generator and discriminator node weightings via back-propagation according to the reward; and
regenerate the graphical workflow elements and re-arranging the elements via the HMC sampling procedure to generate a revised mapping according to the user feedback.
6 . The computer system according to claim 5 , the stored program instructions further comprising program instructions to analyze the feedback according to a dimension selected from the group consisting of tone, sentiment, confusion, and combinations thereof.