🏠 Address Extraction Pipeline

This project extracts addresses from documents using a BERT-based Named Entity Recognition (NER) model.

---

📄 Supported Document Formats

The pipeline supports the following file types:

• ✅ PDF (.pdf)
• ✅ Word Documents (.docx, .doc)
• ✅ Text Files (.txt)

Note: The UnstructuredIO API handles multiple file formats for preprocessing. Additional file types can be supported in future implementations.

---

🔧 Pipeline Workflow

1️⃣ File Preprocessing

• The input document is processed using the UnstructuredIO API.
• Extracted text and metadata are stored in JSON format, which includes:
  • Bounding boxes
  • Page numbers
  • File name
  • Extracted text content

2️⃣ NER Model Processing

• The extracted text is passed through two models:
  • Fine-Tuned BERT Model
  • NER Address Model

3️⃣ Regex-Based Address Validation

• The extracted addresses are checked using a custom regex model.
• Only valid addresses are retained with highest confidence score

4️⃣ Final Address Extraction Output

• The extracted information includes:
  • 📌 Extracted Address
  • 📂 File Number
  • 📄 Page Number
  • 🔲 Approximate Bounding Boxes
  • 🎯 Confidence Score (Maximum value from both models)
  • 🏢 Associated Entity (Organization or Person)

---

📜 Assumptions

• The associated entity (Organization/Person) appears before the address in the document.
• The pipeline assigns the nearest preceding entity as the corresponding entity for each extracted address.

---

📌 Project Structure

• 🏗 Architecture Flow Diagram
  

• 📖 Fine-Tuned Model Notebook (Final Version)
  experimentation/fine_tuning/bert_fine_tuning_v2.ipynb

• 🧪 Test Notebook for Address Extraction Pipeline
  experimentation/address_extraction_v4.ipynb

---

🚀 How to Run the Pipeline

1️⃣ Set Up the Virtual Environment

# Create a Python 3 virtual environment
python3 -m venv env

# Activate the virtual environment
# On Windows:
source env/bin/activate  
# On macOS/Linux:
source env/bin/activate  

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run the Application

streamlit run Home.py

🎯 Using the Streamlit App

• The Home Page provides a summary of the pipeline.
• The Application Tab allows users to upload a file and visualize extracted addresses:
  • 📌 If the file contains addresses, bounding boxes are displayed.
  • 📊 A summary table presents extracted information.