Identify informative words that distinguish between clusters of documents.
Train context vectors in semantic space that can predict words in document, and compare distances.
Subtract frequency of each word in target text at t-1 from frequency in source. Take top n words.
Distinguish topics within source text
Care about target text: content may span whole magazine
Allows substitution of similar words
Context dependent: has to be trained on full range of words to learn similarity
Identify words across entire magazine
Subject to random variation
Compare topic word and topic frequency
Compare distance of context vectors
Compare word frequency