AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

VeriDark: A Large-Scale Benchmark for Authorship Verification on the Dark Web

Andrei Manolache, Florin Brad, Antonio Barbalau, Radu Tudor Ionescu, Marius Popescu

NeurIPS 2022 · Large language models › Language, safety and interpretability · approach: Data and evaluation methodology · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train on Dark-Web-related VeriDark data rather than PAN2020 largeBERT authorship verification · VeriDark AV datasetsavg.: 66.9 -> 77.2; 66.1 -> 84.6; 68.2 -> 88.1Improved
Train on the test-matched dataset rather than a different VeriDark datasetBERT authorship verification · DarkReddit+, SilkRoad1, and Agora AV testsavg.: 77.2; 84.6; 87.3Improved
Train on All VeriDark datasets rather than only a different single datasetBERT authorship verification · Cross-dataset AV testsavg.: 74.7; 83.5; 88.1Improved
Add an Other class using PAN examples during trainingDarkReddit+ author identification · DarkReddit+ test examples, including Otheraccuracy: 84.6 -> 86.5Improved
Add an Other class using PAN examples during trainingDarkReddit+ author identification · ClearReddit test examples, including Otheraccuracy: 81.2 -> 80.4Hurt
Add an Other class from a similar VeriDark sourceDarkReddit+ author identification · DarkReddit+ test examplesaccuracy: 84.6 -> 83.7; 84.6 -> 83.5; 84.6 -> 80.3Hurt
Add an Other class from a VeriDark sourceDarkReddit+ author identification · ClearReddit test examplesaccuracy: 81.2 -> 75.0; 81.2 -> 77.7; 81.2 -> 73.9Hurt
Add reverse chunk pairs during evaluation (ablation)BERT authorship verification · DarkReddit+ AV testavg.: 76.6 -> 77.2Improved
Add reverse chunk pairs during evaluation (ablation)BERT authorship verification · SilkRoad1 AV testavg.: 85.0 -> 85.5Improved
Add reverse chunk pairs during evaluation (ablation)BERT authorship verification · Agora AV testavg.: 87.6 -> 88.0Improved
Pretrain on PAN before VeriDark fine-tuning (ablation)BERT authorship verification · DarkReddit+ AV testavg.: 77.2 -> 77.3No clear effect
Pretrain on PAN before VeriDark fine-tuning (ablation)BERT authorship verification · SilkRoad1 AV testavg.: 84.6 -> 84.2No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy