Ines Panker (2012) Automated authorship attribution for Slovenian literary texts. EngD thesis.
Automatic authorship attribution is an umbrella term for methods trying to derive authorship from text. To achieve this they make use of various data mining techniques. Our chosen task was to test the successfulness of such procedures on a subset of Slovenian literary texts. Each text was represented as a vector with dimensions corresponding to the attributes we decided to measure. We started the calculations by measuring the number of punctuations and continued by measuring the number of word occurrences. We relied on the simple and most known classificators, we tested the SVM, kNN, classification trees and naive Bayes classificator. The last one was found to be giving the best results. Our final results were very satisfactory, with rudimentary approaches we achieved a classification accuracy of 78% and an average precision of 87% with 2 thirds of the authors having precision at 100%.
Actions (login required)