🧠 SkimLit: Medical Abstract Classification

This project focuses on automatic classification of sentences in medical abstracts using different deep learning and machine learning approaches.

We experiment with multiple NLP models ranging from simple baselines to advanced embedding and positional encoding techniques.

---

📌 Problem Statement

Given a medical abstract, classify each sentence into one of the following categories:

• Background
• Objective
• Methods
• Results
• Conclusions

This helps in quick scientific literature understanding (SkimLit task).

---

🎥 YouTube Tutorial Series

The way I learnt project is explained step-by-step in the following video series (In Burmese):

▶️ Part 1: Baseline Model

Introduction to dataset and baseline machine learning model. https://youtu.be/1evvGMdTgY4

---

▶️ Part 2: Convolutional 1D Models

Using Conv1D models for text classification. https://youtu.be/ZmaiD9Y1S0w

---

▶️ Part 3: Embedding Models

Word embeddings and hybrid NLP architectures. https://youtu.be/DupFb4CWGOM

---

▶️ Part 4: Positional Embeddings

Adding positional encoding for better sequence understanding. Connecting...

---

🏗️ Model Approaches Used

This project includes:

• Baseline TF-IDF + classical ML model
• 1D Convolutional Neural Network (Conv1D)
• Word Embedding models (Keras / TensorFlow Hub)
• Hybrid models (token + character inputs)
• Positional embeddings for sequence awareness

---

📁 Project Structure

SkimLit_Medical_Abstract_Classification/
│
├── skimlit_model1.keras # Trained deep learning model
├── app.py # Streamlit web app
├── requirements.txt
├── notebooks/
└── README.md