Visualizing document authorship using n-grams and latent semantic indexing
Ian M. Soboroff, Charles Nicholas, James M. Kukla, David S. Ebert · 1997
An approach to visualizing authorship and writing style of free-form text documents is described. This approach uses n-grams and latent semantic indexing (LSI) to cluster documents according to usage patterns of related n-gram "terms". Latent semantic indexing distributes documents and terms into a relatively low-dimensional space, which can be viewed graphically using various visualization techniques. 1 Introduction Determining the authorship of a text is a common problem that not only occurs in scholarly studies of historical documents, but is also of interest when one seeks to classify large numbers of documents. Knowing the author or authors of a text can provide valuable information on its "context" which might otherwise be unknowable from the text itself. Mathematical techniques have frequently been applied to the problem of document authorship. These approaches analyze such things as frequencies of word usage, sentence structure, and word and sentence length. A typical computa...