Hallucination Detection using Logistic Regression

Overview

This project applies logistic regression for binary text classification, specifically hallucination detection in text summaries. The classifier determines whether a given text is factual or hallucinated based on the XSum Hallucination Dataset.

Dataset

• Source: XSum Hallucination Dataset
• Input: summary field (text summaries)
• Label: is_factual (1 = factual, 0 = hallucinated)

Tasks

🛠 1. Data Preprocessing

• Clean and preprocess text data for model training.

📊 2. Model Training

• Implement logistic regression from scratch (no ML libraries).
• Train the model on the dataset and tune hyperparameters.

📈 3. Model Evaluation

• Assess performance using accuracy, precision, recall, and F1-score.
• Visualize performance using a confusion matrix.

🔄 4. Cross-Validation

• Implement k-fold cross-validation to ensure model robustness.
• Report average accuracy and standard deviation across folds.

🧐 5. Error Analysis

• Identify and analyze misclassified examples.
• Suggest improvements based on findings.

🚀 This project demonstrates logistic regression’s effectiveness in text classification and provides insights into hallucination detection in NLP.