BERT-based Authorship Attribution on the Romanian Dataset called ROST

Abstract

Being around for decades, the problem of Authorship Attribution is still very much in focus currently. Some of the more recent instruments used are the pre-trained language models, the most prevalent being BERT.Here we used such a model to detect the authorship of texts written in the Romanian language. The dataset used is highly unbalanced, i.e., significant differences in the number of texts per author, the sources from which the texts were collected, the time period in which the authors lived and wrote these texts, the medium intended to be read (i.e., paper or online), and the type of writing (i.e., stories, short stories, fairy tales, novels, literary articles, and sketches). The results are better than expected, sometimes exceeding 87% macro-accuracy.

Authorship Attribution in Less-Resourced Languages: A Hybrid Transformer Approach for Romanian

“…Focusing on datasets with 10 authors (see Table 12), we observe that our RoBERT model outperformed the existing approaches [22,51] for both FT and PP corpora. Additionally, our hybrid RoBERT model, which incorporates RBI features, achieved the highest F1 score of 0.95 for the PP corpus, indicating the effectiveness of leveraging both textual and numerical features for AA tasks.…”

“…A second study by Avram [51] focused on authorship attribution using pre-trained language models, particularly BERT, to detect the authorship of Romanian texts. Similar to the previous study, this research used the same dataset, which is highly unbalanced in terms of the number of texts per author, source, time period, and writing type.…”