MarkovQuillClassifier

A Markov Chain-based text classification tool to identify authorship of literary works using n-grams and probabilistic modeling.

---

Overview

MarkovQuillClassifier leverages Markov Chains and N-grams (unigrams and bigrams) to analyze word transitions and patterns in literary texts. It identifies the author of a given text by learning the unique writing style of each author based on transition probabilities.

This project includes:

• N-Gram Features: Extracts unigrams and bigrams for improved context.
• Laplace Smoothing: Handles unseen word transitions.
• Class Balancing: Mitigates class imbalance for better accuracy.
• Performance Metrics: Reports accuracy, confusion matrix, and F1-score for evaluation.

---

Key Features

• Probabilistic Modeling: Uses Markov Chains to model word transitions.
• Custom Vocabulary: Builds vocabulary dynamically with N-grams.
• Performance Evaluation: Outputs accuracy, confusion matrix, and F1-score.
• Lightweight: Built with Python and minimal dependencies.

---

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/MarkovQuillClassifier.git
   cd MarkovQuillClassifier

2. Install required libraries:

   pip install numpy scikit-learn

3. Add your text files (e.g., edgar_allan_poe.txt, robert_frost.txt) in the project directory.

---

Usage

Run the following command to train and evaluate the model:

python nlp_test1.py

Sample Output:

Training Accuracy: 93.73%
Testing Accuracy: 81.90%

Train Confusion Matrix:
[[ 462  108]
 [   0 1153]]

Test Confusion Matrix:
[[ 71  77]
 [  1 282]]

Test F1 Score: 0.8785

---

Applications

• Authorship Attribution: Identify the author of anonymous or disputed texts.
• Text Classification: Classify poems, articles, or literary works.
• Plagiarism Detection: Detect stylistic similarities in text corpora.

---

Future Improvements

• Add support for trigrams to enhance context capture.
• Implement TF-IDF for better word weighting.
• Experiment with advanced models like Naive Bayes or Logistic Regression.
• Expand to classify multiple authors.

---

License

This project is licensed under the MIT License.