Universum Inference and Corpus Homogeneity

Abstract

Abstract. Universum Inference is re-interpreted for assessment of corpus homogeneity in computational stylometry. Recent stylometric research quantifies strength of characterization within dramatic works by assessing the homogeneity of corpora associated with dramatic personas. A methodological advance is suggested to mitigate the potential for the assessment of homogeneity to be achieved by chance. Baseline comparison analysis is constructed for contributions to debates by nonfictional participants: the corpus analyzed consists of transcripts of US Presidential and Vice-Presidential debates from the 2000 election cycle. The corpus is also analyzed in translation to Italian, Spanish and Portuguese. Adding randomized categories makes assessments of homogeneity more conservative.

CLG Authorship Analytics: a library for authorship verification

“…Various methods have been proposed in the literature to address author verification. In particular ensemble methods have proved quite sucessful at tackling the challenge of capturing the features relevant to the author's style and discarding the ones which are not [7][8][9][10]21]. Author verification has also been the focus of several iterations of the PAN shared tasks, 4 e.g.…”

“…A common option is to use the results of some Google queries formed by randomly picking words from the set of input documents as impostors. 8 In the experiments presented below (see section 5), we opt for using all the training documents as impostors. While this option is not ideal since the documents obviously include precisely the documents to be compared, it is a reasonable simplification if the training set is diverse enough in terms of authors and if the number of iterations is large enough to prevent the occasional wrong comparison from having a significant effect on the output features.…”

“…This strategy follows the idea described in [8]: in this paper, a large corpus containing several "categories" (the input documents in our task) is split into small chunks. A chunk of category A is compared against many chunks from other categories and from category A as well, picked randomly.…”

“…3. The three categories are compared in the same way as in [8], that is, both against itself (using the two parts belonging to this category) and against each other category (picking one of the two parts randomly).…”

Found in translation

“…It is convenient to conflate the notions of "methods" and "tools". Various aspects of the tools and analysis conducted using the tools have been published (Appel & Vogel, 2001;Van Gijsel & Vogel, 2003;O'Brien & Vogel, 2003;Vogel, 2007b;Healey, Vogel, & Eshghi, 2007;Vogel, 2007a;Vogel & Brisset, 2007;Frontini, Lynch, & Vogel, 2008;Vogel, Lynch, & Janssen, 2008). Student projects (e.g.…”