Subtopic Deep Dive

Adversarial Stylometry and Obfuscation
Research Guide

What is Adversarial Stylometry and Obfuscation?

Adversarial stylometry studies attacks on authorship attribution systems through style manipulation, while obfuscation develops countermeasures to hide author identity without altering semantics.

Brennan et al. (2012) introduced adversarial stylometry, demonstrating how authors can evade detection with 186 citations. Afroz et al. (2012) explored deception detection in online writing, achieving 333 citations. Over 10 papers from 2009-2023 benchmark robustness against GANs, paraphrasing, and neural translation obfuscation.

15
Curated Papers
3
Key Challenges

Why It Matters

Adversarial stylometry counters deanonymization threats in digital forensics and social media, as Brennan and Greenstadt (2009) showed practical attacks reducing attribution accuracy by over 50% (101 citations). Reddy and Knight (2016) obfuscated gender signals in tweets, enabling privacy in targeted advertising (93 citations). Shetty et al. (2023) used adversarial NMT training for attribute anonymity, protecting against age/gender inference in anonymous posts (68 citations). Applications span criminal investigations, academic integrity like Ison (2020) contract cheating detection (43 citations), and secure AI communications.

Key Research Challenges

Evasion Effectiveness

Attackers manipulate features like function words to fool SVM classifiers, as Brennan et al. (2012) reduced accuracy from 90% to 20% (186 citations). Balancing semantic preservation remains hard. Defenses lag behind evolving attacks.

Closed-World Assumption

Models fail on unseen authors, per Stolerman et al. (2014) breaking assumptions with open-set attacks (29 citations). Scalability to millions of candidates challenges efficiency. Real-world data scarcity hinders training.

Attribute Obfuscation

Hiding gender or age via paraphrasing succeeds partially, as Reddy and Knight (2016) showed (93 citations), but neural detectors adapt quickly. Shetty et al. (2023) adversarial training helps but increases compute (68 citations). Generalization across languages is limited.

Essential Papers

1.

Detecting Hoaxes, Frauds, and Deception in Writing Style Online

Sadia Afroz, Michael Brennan, Rachel Greenstadt · 2012 · 333 citations

In digital forensics, questions often arise about the authors of documents: their identity, demographic background, and whether they can be linked to other documents. The field of stylometry uses l...

2.

Adversarial stylometry

Michael Brennan, Sadia Afroz, Rachel Greenstadt · 2012 · ACM Transactions on Information and System Security · 186 citations

The use of stylometry, authorship recognition through purely linguistic means, has contributed to literary, historical, and criminal investigation breakthroughs. Existing stylometry research assume...

3.

When Coding Style Survives Compilation: De-anonymizing Programmers from Executable Binaries

Aylin Caliskan, Fabian Yamaguchi, Edwin Dauber et al. · 2018 · 105 citations

The ability to identify authors of computer programs based on their coding\nstyle is a direct threat to the privacy and anonymity of programmers. While\nrecent work found that source code can be at...

4.

Practical Attacks Against Authorship Recognition Techniques

Michael Brennan, Rachel Greenstadt · 2009 · 101 citations

Abstract.The use of statistical AI techniques in authorship recognition (or stylometry) has contributed to literary and historical breakthroughs. These successes have led to the use of these techni...

5.

Obfuscating Gender in Social Media Writing

Sravana Reddy, Kevin Knight · 2016 · 93 citations

The vast availability of textual data on social media has led to an interest in algorithms to predict user attributes such as gender based on the user's writing.These methods are valuable for socia...

6.

A4NT: Author Attribute Anonymity by Adversarial Training of Neural Machine Translation

Rakshith Shetty, Bernt Schiele, Mario Fritz · 2023 · MPG.PuRe (Max Planck Society) · 68 citations

Text-based analysis methods enable an adversary to reveal privacy relevant author attributes such as gender, age and can identify the text's author. Such methods can compromise the privacy of an an...

7.

Machine Learning for Ancient Languages: A Survey

Thea Sommerschield, Yannis Assael, John Pavlopoulos et al. · 2023 · Computational Linguistics · 53 citations

Abstract Ancient languages preserve the cultures and histories of the past. However, their study is fraught with difficulties, and experts must tackle a range of challenging text-based tasks, from ...

Reading Guide

Foundational Papers

Start with Afroz et al. (2012, 333 citations) for deception basics, then Brennan et al. (2012, 186 citations) for attacks, Brennan and Greenstadt (2009, 101 citations) for practical evasions.

Recent Advances

Study Shetty et al. (2023) for NMT anonymity (68 citations), Ison (2020) cheating detection (43 citations), El-Fiqi et al. (2019) motifs (40 citations).

Core Methods

Core techniques: SVM on function words (Brennan 2012), network automata (Machicao et al., 2018), adversarial training (Shetty 2023), paraphrasing (Reddy 2016).

How PapersFlow Helps You Research Adversarial Stylometry and Obfuscation

Discover & Search

Research Agent uses citationGraph on Brennan et al. (2012) to map 186-cited adversarial stylometry works, then findSimilarPapers uncovers Shetty et al. (2023) NMT obfuscation. exaSearch queries 'authorship obfuscation GANs' for 250M+ OpenAlex papers beyond provided lists.

Analyze & Verify

Analysis Agent runs readPaperContent on Afroz et al. (2012), verifies deception metrics via runPythonAnalysis replicating SVM feature extraction with NumPy/pandas, and applies GRADE grading to classify evidence strength. verifyResponse (CoVe) chain checks statistical claims against Brennan and Greenstadt (2009) attack accuracies.

Synthesize & Write

Synthesis Agent detects gaps in obfuscation defenses post-Reddy and Knight (2016), flags contradictions between 2012 Brennan attacks and 2023 Shetty advances. Writing Agent uses latexEditText for obfuscation algorithm pseudocode, latexSyncCitations for 10+ papers, latexCompile for report, exportMermaid for attack-defense flow diagrams.

Use Cases

"Replicate Brennan 2012 adversarial stylometry attack success rates on modern SVMs"

Research Agent → searchPapers 'Brennan adversarial stylometry' → Analysis Agent → runPythonAnalysis (pandas SVM retrain on extracted features) → GRADE-verified accuracy plot.

"Draft LaTeX review of obfuscation methods from Afroz to Shetty"

Research Agent → citationGraph (Brennan 2012 cluster) → Synthesis → gap detection → Writing Agent → latexEditText (intro/methods) → latexSyncCitations → latexCompile PDF.

"Find GitHub repos implementing Reddy 2016 gender obfuscation"

Research Agent → paperExtractUrls (Reddy Knight) → Code Discovery → paperFindGithubRepo → githubRepoInspect (code review, runPythonAnalysis on stylometric eval scripts).

Automated Workflows

Deep Research scans 50+ stylometry papers via searchPapers → citationGraph, outputs structured report on evasion trends from Brennan (2009-2012) to Shetty (2023). DeepScan's 7-step chain analyzes Juola (2011) obfuscation with readPaperContent → CoVe verification → runPythonAnalysis benchmarks. Theorizer generates hypotheses on GAN-resistant stylometers from Afroz et al. (2012) deception patterns.

Frequently Asked Questions

What is adversarial stylometry?

Adversarial stylometry develops attacks disguising author style, as Brennan et al. (2012) perturbed features to evade detectors (186 citations).

What methods obfuscate authorship?

Methods include feature removal (Brennan and Greenstadt, 2009), paraphrasing (Reddy and Knight, 2016), and adversarial NMT (Shetty et al., 2023).

What are key papers?

Foundational: Afroz et al. (2012, 333 citations), Brennan et al. (2012, 186 citations). Recent: Shetty et al. (2023, 68 citations), Ison (2020, 43 citations).

What open problems exist?

Open-set attribution beyond closed worlds (Stolerman et al., 2014), cross-lingual obfuscation, and scalable defenses against adaptive neural attacks.

Research Authorship Attribution and Profiling with AI

PapersFlow provides specialized AI tools for Computer Science researchers. Here are the most relevant for this topic:

See how researchers in Computer Science & AI use PapersFlow

Field-specific workflows, example queries, and use cases.

Computer Science & AI Guide

Start Researching Adversarial Stylometry and Obfuscation with AI

Search 474M+ papers, run AI-powered literature reviews, and write with integrated citations — all in one workspace.

See how PapersFlow works for Computer Science researchers