🧠 Patient Sentiment Analysis Using NLP & LLMs

This project analyzes open-ended patient reviews to predict sentiment (positive or negative) using Natural Language Processing (NLP) techniques and Large Language Models (LLMs).

🚀 Project Overview

• Goal: Classify patient feedback based on sentiment
• Dataset: 996 hospital reviews with labeled sentiment
• Techniques used:
  • Text cleaning and preprocessing (NLTK, regex)
  • Exploratory Data Analysis (word clouds, word frequencies, review length)
  • Feature extraction with TF-IDF
  • Sentiment classification using:
    • Logistic Regression (baseline)
    • distilBERT LLM from Hugging Face (zero-shot)

🗂️ Project Structure

patient-sentiment-healthcare/
├── data/
│ ├── dataset_hospital_reviews.csv #raw
│ └── dataset_hospital_reviews_cleaned.csv #processed
├── notebooks/
│ ├── 01_data_cleaning_and_eda.ipynb # Data cleaning + EDA
│ └── 02_modeling_and_llm_comparison.ipynb # Model training + LLM comparison
├── README.md

📊 Results Summary

Logistic Regression (TF-IDF)

• Accuracy: 0.86
• High precision on positive class
• Poor recall on negative class

distilBERT (LLM)

• Accuracy: 0.78
• Much better at identifying negative reviews
• Balanced recall across classes

🧪 Example Review

"Wait hour despite appointment isn’t first time happened understanding manage appointment queue it’s random unorganised lot scope improve"

--> Detected as NEGATIVE by distilBERT

🛠️ Tech Stack

• Python, Pandas, Scikit-learn, NLTK, Matplotlib, Seaborn
• Hugging Face Transformers (distilBERT)
• Google Colab (for LLM execution)

📁 How to Run

1. Open 02_modeling_and_llm_comparison.ipynb in Google Colab
2. Mount your Google Drive and upload the cleaned dataset (or use the one provided)
3. Run the cells to explore, train, and evaluate both models