Google Patent About Author Vectors to Identify Author of a Text

Google was recently granted a patent called Generating author vectors that uses a neural network system and an author vector to characterize a particular author.

A very generalized version of it is that it starts with obtaining a set of sequences of words. That set of sequences of words make up a number of first sequences of words. For each of those first sequences of words, the second sequence of words follows that first sequence of words. That first sequence of words and each second sequence of words can be classified as being authored by a particular author. A neural network system could be trained on those sets of words to determine an author, and an author vector may be used to characterize a particular author.

An author vector that effectively characterizes an author can be generated from a text written by the author without that text being labelled. Once generated, the author vector can characterize different properties of the author depending on the context of the use of the author vector. By clustering the author vectors, clusters of authors that have similar communication styles and, in some implementations, personality types can be effectively be generated.