Authorship identification using ensemble learning.

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• The proliferation of digital text necessitates effective methods for author identification.
• Current authorship attribution techniques face challenges with large, anonymous datasets.

Purpose of the Study:

• To develop and evaluate a novel authorship identification system.
• To enhance the accuracy of identifying authors from textual data using advanced machine learning models.

Main Methods:

• Utilized ensemble learning, DistilBERT, and conventional machine learning.
• Employed count vectorizer and bi-gram Term Frequency-Inverse Document Frequency (TF-IDF) for feature extraction.
• Experimented on the "All the news" dataset with 10 and 20 authors.

Main Results:

• The proposed ensemble learning and DistilBERT approaches demonstrated superior performance across all dataset subsets.
• Achieved accuracy gains of up to 5.25% with ensemble learning and 7.17% with DistilBERT for 20 authors.
• Results indicate significant improvements over existing state-of-the-art methods in authorship identification.

Conclusions:

• The novel approach effectively identifies authors from unknown texts.
• Ensemble learning and DistilBERT offer a promising direction for advanced authorship attribution.
• The system provides a robust solution for the growing challenge of anonymous digital content.