Automated Authorship Attribution using CNG Distance on Blog Posts in the Serbian Language
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The automated authorship attribution problem is a task of identifying the author of a given text using an objective algorithmic method based on previous texts written by the candidate authors. We are particularly interested in methods that do not rely on any language-specific knowledge or preprocessing, and that are based on a low-level text representation such as a sequence of letters and other characters. The previous work has shown that author profiles consisting of the most frequent character n-grams are effective in the authorship attribution in a number of languages, but not many results are reported on languages with sparse resources, such as the Serbian and related languages. We show that a character n-gram based method has also a very good performance in the Serbian language. Another contribution of this work is a new dataset prepared as a good benchmark for the authorship attribution task, comparable to the previously published similar datasets for English, Greek, and some other languages. This dataset for authorship attribution prepared in this work consists of blog posts published as commentary columns on a news and commentary portal, and as such is a grammatical and well-written language corpus, and a good representative of current normative language. The CNG distance method, which was shown to work well in a number of languages before, shows high accuracy of 94% over 5 authors, and 83% over 10 authors in the authorship attribution for this dataset as well. As expected from the results for other European languages, the highest accuracy is obtained around n-grams of size n =6,7, or a wider range of n =3,…,8, with L parameter from 500 to 9000, although even for the parameters n =2 and L =500 some relatively high accuracies are achieved.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it