Plot Arceology: a vector-space model of narrative structure

摘要

A novel and important corpus of about 80,000 television and movie scripts from opensubtitles.com shows interesting large-scale patterns of narration in their vocabulary use. These patterns are interesting at the token level but not easily amenable for large scale data analysis. This paper describes a new method, "plot arcs," for describing and comparing structural elements of structure, including plot, across large textual corpora by treating texts as paths through a multidimensional space derived from a topic model. Plot arcs offer a framework for describing the structure of text documents that is easily extensible to a variety of genres and can accommodate many different ideas of plot structure.