An ensemble deep learning model for author identification through multiple features.

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• Authorship identification is a challenging task in natural language processing.
• Existing methods often struggle with accuracy and robustness across diverse writing styles.

Purpose of the Study:

• To develop a novel deep learning framework for improved authorship identification.
• To enhance the accuracy and stability of author identification models.

Main Methods:

• A self-attentive weighted ensemble framework combining various feature types (statistical, TF-IDF, Word2Vec).
• Separate Convolutional Neural Networks (CNNs) for extracting specific stylistic features.
• A self-attention mechanism to dynamically weigh the importance of different feature sets.
• A weighted SoftMax classifier for optimized performance.

Main Results:

• The proposed model achieved 80.29% accuracy on Dataset A (4 authors) and 78.44% on Dataset B (30 authors).
• Outperformed state-of-the-art methods by 3.09% on Dataset A and 4.45% on Dataset B.
• Demonstrated significant gains in accuracy and robustness for author identification.

Conclusions:

• The self-attention-augmented multi-feature ensemble approach is highly effective for authorship identification.
• The framework shows strong generalization capabilities by integrating diverse writing style representations.
• This research offers a robust solution for the complex problem of author attribution.