Joint understanding of actors, literary characters, and movies

A computer-implemented method, comprising:

• receiving information specifying at least movies, keywords describing the movies, and actors appearing in the movies;
• initializing means and co-variance matrices of a plurality of Gaussian distributions representing the movies, keywords, and actors with random values;
• and
• optimizing, based at least in part on the received information, the plurality of Gaussian distributions.

A computer-implemented method, comprising:

• receiving text describing one or more movies and characters therein;
• performing coreference resolution to link pronouns in the received text with the characters;
• identifying words in the received text associated with actions performed by the characters, actions received by the characters, and descriptions of the characters;
• mapping the identified words associated with the actions performed by the characters, the actions received by the characters, and the descriptions of the characters to numerical representations;
• averaging the numerical representations of the words associated with the actions performed by the characters, the actions received by the characters, and the descriptions of the characters;
• concatenating, for each character, the averaged numerical representations associated with the character into a vector representing the character;
• identifying archetypical roles as clusters of the vectors representing the characters;
• and
• training a model based, at least in part, on the identified archetypical roles.