Deep learning of bots through examples and experience

A method comprising:

• determining marked-up text from an input text, wherein the marked-up text associates portions of the input text with corresponding entities identified in the text;
• identifying available actions, from among a plurality of available actions based on the identified entities in the input text;
• providing the identified available actions and a set of features to a recurrent neural network, the set of features including at least one feature associated with a previous action if the previous action took place;
• receiving an indication of a selected action from the recurrent neural network;
• updating an internal state based on the indication of a selected action;
• and,
• initiating the selected action, including one of generating textual output, invoking an application programming interface (API) call, receiving input from a user, and quitting an interaction with the user.

A system comprising:

• one or more processors;
• and,
• memory in communication with the one or more processors, the memory comprising code, that when executed, causes the one or more processors to control the system to:
• receive one or more sample dialogs;
• determine if any action among a plurality of actions does not correspond to any entity among the one or more sample dialogs;
• and in response to a determination that every action in the plurality of actions corresponds to an entity among the one or more sample dialogs:
• incorporate all of the one or more sample dialogs into a training set;
• create a log including a set of features and entity extractions;
• apply training using the training set and the log in a recurrent neural network;
• score the one or more sample dialogs using the recurrent neural network to generate a score result based on the training;
• and,
• determine if a target action in the one or more sample dialogs was not assigned a highest probability in the score result.