Victorian Author Attribution using Machine Learning

Authors: Tom BOISSE & Jules ESPINOUX

---

Project Overview

This project implements machine learning algorithms to predict authorship of text excerpts from Victorian-era novelists. Using the Gungor 2018 Victorian Author Attribution dataset, we train and evaluate various classifiers to identify which of 50 different authors wrote a given text passage.

The main challenge addressed is transforming unstructured text data into meaningful numerical representations while preserving linguistic patterns and stylistic features that characterize each author's writing.

---

Objectives

• Authorship Attribution: Predict which author wrote a text excerpt from 50 Victorian-era novelists
• Text Vectorization: Compare different text representation methods (Bag-of-Words, TF-IDF)
• Classification: Evaluate multiple ML algorithms (Naive Bayes, SVM) for this multi-class classification problem
• Author Analysis: Extract and visualize statistical features of different authors writing styles

---

Dataset

Source

Gungor, A. (2018). Fifty Victorian Era Novelists Authorship Attribution Data. IUPUI University Library.
DOI: http://dx.doi.org/10.7912/D2N65J

---

Project Structure

ML-Authors/
├── CR_be_ML_BOISSE_ESPINOUX.ipynb    # Main notebook with complete analysis
├── main.py                            # Quick model training script
├── tokenizer.py                       # Custom tokenization utilities
├── bow.py                             # Bag-of-Words implementation
├── stats.py                           # Generate author statistics
├── plot_stats.py                      # Visualize author statistics
├── dataset/                           # Data files
├── tokenizer.txt                      # Word-to-index mapping
├── stats.txt                          # Computed author statistics
├── Data Description.pdf               # Additional dataset documentation
└── README.md                          # This file

---

Full Analysis

Open and run the Jupyter notebook for complete analysis:

jupyter notebook CR_be_ML_BOISSE_ESPINOUX.ipynb

The notebook includes:

• Data exploration and visualization
• Multiple vectorization methods (BoW, TF-IDF)
• Multiple classifier comparisons (Naive Bayes, SVM)
• Hyperparameter tuning
• Results analysis and discussion

---

Methodology

1. Text Vectorization

We implemented and compared two approaches:

Bag-of-Words (BoW)

• Counts word occurrences in each text
• Creates sparse matrix representation
• Each dimension represents a unique word
• Custom implementation in bow.py
• Scikit-learn implementation via CountVectorizer

TF-IDF (Term Frequency - Inverse Document Frequency)

• Weights words by importance
• TF: frequency of word in document
• IDF: inverse frequency across all documents
• Implemented via TfidfVectorizer

2. Classification Algorithms

Naive Bayes (MultinomialNB)

• Probabilistic classifier
• Assumes feature independence
• Fast training and prediction
• Well-suited for text classification

Support Vector Machine (LinearSVC)

• Finds optimal hyperplane separating classes
• Effective in high-dimensional spaces
• Good generalization with proper regularization
• Hyperparameter: C (regularization strength)

---

Key Findings

(Based on typical performance patterns for this dataset)

• TF-IDF generally outperforms simple BoW
• Linear SVC achieves competitive accuracy
• Proper hyperparameter tuning (C parameter) significantly impacts performance
• Some authors are easier to distinguish than others due to distinct writing styles

---

Technical Details

Data Preprocessing

• Dataset is pre-normalized (no additional preprocessing needed)
• Text already cleaned and standardized
• Consistent 1000-word excerpts

Feature Engineering

• Vectorizers trained on entire corpus for better vocabulary coverage
• Word frequency-based tokenization
• Sparse matrix representation for memory efficiency

Implementation Highlights

• Scikit-learn pipelines for clean model composition
• Custom BoW implementation using scipy.sparse.csr_matrix
• Efficient word counting with collections.Counter

---