Stylometry

This repository contains Python scripts for performing stylometric analysis using various techniques:

1. burrows-delta-mds.py: Computes Burrows's Delta and visualises the results using Multidimensional Scaling (MDS).
2. burrows-delta-dendrogram.py: Computes Burrows's Delta and visualises the results using a dendrogram based on hierarchical clustering.
3. roberta-embeddings.py: Analyses stylistic similarities among literary texts using the RoBERTa model for embedding generation, dimensionality reduction via PCA, and visualisation through scatter plots.

Scripts Overview

1. burrows-delta-mds.py

Purpose: Computes Burrows's Delta and visualises the results using Multidimensional Scaling (MDS).

Output: A scatter plot where points represent texts, and their proximity reflects stylistic similarity.

Features:

• Most Frequent Words (MFW): Set to 100 by default.
• Saves the Burrows's Delta matrix as burrows_delta_matrix.csv.
• Saves the MDS visualisation as mds_visualisation.png.

Usage:

Run the script in your terminal:

python3 burrows-delta-mds.py

Dependencies:

• nltk
• pandas
• numpy
• matplotlib
• scikit-learn

2. burrows-delta-dendrogram.py

Purpose: Computes Burrows's Delta and visualises the results using a dendrogram based on hierarchical clustering.

Output: A dendrogram where labels are colour-coded by group, extracted from the text before the first _ in each filename.

Features:

• Linkage Method: Uses average linkage (default in stylometry for balanced clustering).
• Colour-Coded Labels: Groups are derived from the filenames (e.g., group_filename.txt).
• Saves the Burrows's Delta matrix as burrows_delta_matrix.csv.
• Saves the dendrogram visualisation as dendrogram_visualisation_coloured.png.

Usage:

Run the script in your terminal:

python3 burrows-delta-dendrogram.py

Dependencies:

• nltk
• pandas
• numpy
• matplotlib
• scipy

3. roberta-embeddings.py

Purpose: Analyses stylistic similarities among literary texts using the RoBERTa model for embedding generation, dimensionality reduction via PCA, and visualisation through scatter plots.

Input Requirements

Corpus

Both burrows-delta scripts expect a folder named corpus containing .txt files. Each file should represent a single text.

For burrows-delta-dendrogram.py, filenames should follow the format:

<group>_rest_of_filename.txt

For roberta-embeddings.py, the folder should be named lit-families on your Desktop. Filenames should follow the structure:

surname_firstinitial_title.txt

Example Folder Structure

corpus/
├── group1_text1.txt
├── group1_text2.txt
├── group2_text1.txt
├── group2_text2.txt

lit-families/
├── Joyce_J_Ulysses.txt
├── Woolf_V_ToTheLighthouse.txt

Outputs

Common Outputs

• Delta Matrix: Saved as burrows_delta_matrix.csv. A symmetric matrix of stylistic distances between texts.

Script-Specific Outputs

1. MDS Script (burrows-delta-mds.py):

   • Scatter Plot: Saved as mds_visualisation.png.

2. Dendrogram Script (burrows-delta-dendrogram.py):

   • Dendrogram Plot: Saved as dendrogram_visualisation_coloured.png.

3. RoBERTa Script (roberta-embeddings.py):

   • Scatter plots are displayed in separate windows but are not saved automatically.

Customisation

Adjusting Most Frequent Words (MFW)

Both burrows-delta scripts use the top 100 Most Frequent Words (MFW) by default. To change this, modify the mfw parameter in the compute_frequencies function:

frequency_matrix = compute_frequencies(preprocessed_texts, mfw=200)  # Example: Use 200 MFW

Changing Linkage Method in Dendrogram

The dendrogram script uses average linkage by default. To change the method, update the linkage function:

linkage_matrix = linkage(condensed_matrix, method='complete')  # Use complete linkage

Available methods: 'single', 'complete', 'average', 'ward'.

RoBERTa Visualisation Saving

To save the visualisations from roberta-embeddings.py, modify the plt.show() lines to include saving functionality, such as:

plt.savefig('plot1.png')

Dependencies

Install required Python packages using pip:

pip install nltk pandas numpy matplotlib scipy scikit-learn sentence-transformers

How to Run Scripts

1. Place the respective script in any directory on your machine.
2. Ensure the required folder (corpus or lit-families) exists and contains .txt files.
3. Run the script using:

python <script_name>.py

4. View the generated plots or outputs.

Notes

• Model Choice: roberta-embeddings.py uses all-roberta-large-v1, a RoBERTa-based transformer model trained for sentence embeddings.
• Customisation: You can replace the model with any SentenceTransformer model. Update the following line:

model = SentenceTransformer('all-roberta-large-v1')

• Text Preprocessing: Minimal preprocessing is applied. Add additional steps if needed for your corpus.

Troubleshooting

1. No Plots Displayed:

   • Ensure matplotlib is installed.
   • Check for errors during PCA or embedding generation.

2. Incorrect Labels:

   • Verify the file naming structure. Non-conforming files are labelled as "Unknown."