Authorship Attribution (Stylometry)
Also known as: Stylometry, Authorship Analysis, Yazarlık Atıfı, Authorship Identification
Authorship attribution is the task of identifying the most probable author of an anonymous or disputed text by analysing its stylistic fingerprint. Rooted in the statistical work of Mosteller and Wallace on the Federalist Papers (1964), the field was systematically surveyed and formalised by Stamatatos (2009), who catalogued feature sets ranging from character n-grams and function-word frequencies to syntactic and semantic representations used by modern machine-learning classifiers.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use authorship attribution when you have labelled training samples from each candidate author and need to assign an anonymous or disputed document to one of them. The method assumes stylistic consistency within an author and sufficient stylistic divergence between authors. It works best with texts of at least a few hundred words; very short texts (tweets, single paragraphs) yield unreliable features. When no candidate pool exists, consider authorship verification or clustering instead.
Strengths & limitations
- Handles large candidate sets efficiently using profile-based centroid or classifier approaches.
- Character n-gram and function-word features are topic-agnostic and highly discriminative.
- Well-studied benchmark datasets (PAN shared tasks) enable reproducible evaluation.
- Scalable to cross-lingual settings with language-independent feature types.
- Requires sufficient training text per author — sparse corpora degrade accuracy substantially.
- Performance drops on very short texts where stylistic signals are statistically unreliable.
- Deliberate stylistic imitation or obfuscation can fool feature-based classifiers.
- Closed-set assumption fails in open-set scenarios where the true author may not be among candidates.
Frequently asked
How much training text is needed per author?
The minimum varies by feature type and the number of candidates, but empirical studies suggest at least 5,000–10,000 words per author for reliable character n-gram profiles. With fewer words, cross-validation estimates become unstable and error rates rise sharply, particularly when candidate counts exceed ten authors.
Can the method handle completely unknown authors (open-set attribution)?
Standard closed-set attribution assumes the true author is among the candidates, so it will always assign a label. Open-set attribution adds a rejection threshold: if no candidate profile achieves a similarity above the threshold, the document is flagged as 'unknown author'. Calibrating this threshold requires a held-out set of documents from outside the candidate pool.
Are neural language model embeddings better than traditional stylometric features?
Recent work shows that fine-tuned transformer embeddings (e.g., BERT) can outperform handcrafted features on same-domain benchmarks, but they are more sensitive to topical drift and require more labelled data. Character n-gram and function-word baselines remain competitive in low-resource and cross-genre settings and are far more interpretable for forensic reporting.
Sources
- Stamatatos, E. (2009). A survey of modern authorship attribution methods. Journal of the American Society for Information Science and Technology, 60(3), 538–556. DOI: 10.1002/asi.21001 ↗
How to cite this page
ScholarGate. (2026, June 2). Authorship Attribution (Stylometry). ScholarGate. https://scholargate.app/en/text-mining/authorship-attribution
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Forensic Likelihood RatioForensic Science↔ compare
- Text ClassificationText mining↔ compare
- Word2VecText mining↔ compare