Digital Authorship Attribution in Russian-Language Fanfiction and Classical Literature

This article is the third paper in a series aimed at establishment of authorship Russian-language texts. considers methods for determining classical Russian literary texts, as well fanfiction The process author was first considered version classification experiments using closed set authors, and were also completed complicated modification problem an open authors. use to identify text justified by conclusions about effectiveness fastText Support Vector Machine (SVM) with selection informative features discussed our past studies. In case attribution, proposed are based on author’s combination One-Class SVM statistical estimates vector’s similarity measures. feature algorithm authors chosen comparison five different methods, including previously genetic baseline. regularization-based (RbFS) found be most efficient method, while complete enumeration (FFS SFS) ineffective any accuracy RbFS texts averaged 83%, which outperforms other 3 10% identical number features, average 84%. For attribution cross-topic classification, method 85%, in-group it 75 78%, depending group, best result among considered.