EFAuR

Embeddings For Authorship using RoBERTa

• EFAuR
• Dataset
  • Dataset Collection Process
• Models
  • Activations and Definitions

Dataset

The dataset used in this project consists of public domain books written in English, with a single, known author, sourced from Project Gutenberg.

Dataset Collection Process

1. Mirror all .txt files from a Project Gutenberg Mirror
2. Download the metadata tarball from gutenberg.org
3. Extract the metadata tarball and parse all the RDF metadata files within
   • Select English books labeled for public domain use with a single, known author
   • Write the ID of each book, along with the corresponding author, to a CSV file
4. Combine all books written by an author into one file
   • Combine all books by the same author into a single file
   • Remove Project Gutenberg headers and footers
   • Remove readability formatting that may obscure the texts
5. Create a dataset directory
   • Create subdirectories for train, test, and val
   • Each subdirectory has files containing a single data point for the model
   • Each subdirectory has text1, text2, and a label specifying same or different authorship
     • label is 0 if same authorship, 1 if different

Models

Activations and Definitions