Let x represent an unknown document and let y represent per random target author’s stylistic ‘profile’. During one hundred iterations, it will randomly select (a) fifty per cent of the available stylistic features available (anche.g. word frequencies) and (b) thirty distractor authors, or ‘impostors’ from verso pool of similar texts. Per each iteration, the GI will compute whether interrogativo is closer esatto y than preciso any of the profiles by the thirty impostors, given the random selection of stylistic features durante that iteration. Instead of basing the verification of the direct (first-order) distance between quantitativo and y, the GI proposes preciso superiorita the proportion of iterations mediante which quantita was indeed closer esatto y than puro one of the distractors sampled. This proportion can be considered per second-order metric and will automatically be verso probability between zero and one, indicating the robustness of the identification of the authors of quantitativo and y. Our previous work has already demonstrated that the GI system produces excellent verification results for classical Latin prose.31 31 Amico the setup con Stover, et al, ‘Computational authorship verification method’ (n. 27, above). Our verification code is publicly available from the following repository: This code is described durante: M. Kestemont et al. ‘Authenticating the writings’ (n. 29, above).
For modern documents, Koppel and Winter were even able onesto report encouraging scores for document sizes as small as 500 words
We have applied a generic implementation of the GI esatto the HA as follows: we split the individual lives into consecutive samples of 1000 words (i.ed. space-free strings of alphabetic characters), after removing all punctuation.32 32 Previous research (see the publications mentioned durante the previous two notes) suggests that 1,000 words is per reasonable document size con this context. Each of these samples was analysed individually by pairing it with the profile of one of the HA’s six alleged authors, including the profile consisting of the rest of the samples from its own text. We represented the sample (the ‘anonymous’ document) by verso vector comprising the relative frequencies of the 10,000 most frequent tokens per the entire HA. For each author’s profile, we did the same, although the profile’s vector comprises the average correspondante frequency of the 10,000 words. Thus, the profiles would be the so-called ‘mean centroid’ of all individual document vectors for a particular author (excluding, of course, the current anonymous document).33 33 Koppel and Seidman, ‘Automatically identifying’ (n. 30, above). Note that the use of verso scapolo centroid a author aims onesto scampato, at least partially, the skewed nature of our momento, since some authors are much more strongly represented sopra the corpo or retroterra pool than others. If we were not using centroids but mere text segments, they would have been automaticallysampled more frequently than others during the imposter bootstrapping.
Sicuro the left, verso clustering has been added on culmine of the rows, reflecting which groups of samples behave similarly
Next, we ran the verification approach. During one hundred iterations, we would randomly select 5,000 of the available word frequencies. We would also randomly sample thirty impostors from verso large ‘impostor pool’ of documents by Latin authors, including historical writers such as Suetonius and Livy.34 34 See Appendix 2 for the authors sampled. The pool of impostor texts can be inspected mediante the code repository for this paper. Sopra each iteration, we would check whether the anonymous document was closer to the current author’s profile than esatto any of the impostors sampled. In segno mingle2 in this study, we use the ‘minmax’ metric, which was recently introduced durante the context of the GI framework.35 35 See Koppel and Winter, ‘Determining if two documents’ (n. 26, above). For each combination of an anonymous text and one of the six target authors’ profiles, we would record the proportion of iterations (i.anche. a probability between zero and one) sopra which the anonymous document would indeed be attributed esatto the target author. The resulting probability table is given sopra full per the appendix preciso this paper. Although we present verso more detailed dialogue of this scadenza below, we have added Figure 1 below as an intuitive visualization of the overall results of this approach. This is a heatmap visualisation of the result of the GI algorithm for 1,000 word samples from the lives durante the HA. Cell values (darker colours mean higher values) represent the probability of each sample being attributed preciso one of the alleged HA authors, rather than an imposter from verso random selection of distractors.