📚 Autonomous Realtime Research Paper Classifier with Retrieval-Augmented Generation (RAG) using Pathway Services

🌟 Overview

This project is a comprehensive research paper classification system developed during the Kharagpur Data Science Hackathon 2025, where our team emerged as one of the top 10 teams out of 10,000 participants after three intense rounds of competition.

Machine Learning, Retrieval-Augmented Generation (RAG), and an interactive Streamlit-based UI are combined to classify research papers into their target conferences while providing detailed reasoning for the classification. 🚀

---

✨ Key Features

1. 🎛️ Streamlit-Based User Interface

   • Upload research paper PDFs via an intuitive web application.
   • Receive real-time conference classification and detailed rationale.

2. 🤖 Machine Learning Classification

   • Trained classifiers like Logistic Regression, SVM, and Naive Bayes for initial publishability checks.
   • Used a voting classifier for robust predictions.

3. 🔍 Retrieval-Augmented Generation (RAG)

   • Integrated Pathway’s Vector Store for semantic similarity search, dynamically retrieving relevant reference papers.
   • Utilized GeminiEmbedder for creating high-quality embeddings and TokenCountSplitter for efficient chunking.
   • Enabled real-time updates of reference papers via the Google Drive Connector.

4. 🧠 AI-Powered Conference Classification

   • Integrated Gemini AI to classify papers and generate explanations (rationale) for transparency and accuracy.

5. ⚙️ Multithreaded Architecture

   • Developed a multithreaded Vector Store Server to ensure seamless querying and retrieval operations without interruptions.

---

🛠️ Technologies and Skills

🔑 Key Skills:

• Python: Backend programming for machine learning models and preprocessing.
• Streamlit: User-friendly web-based application development.
• Machine Learning: Ensemble voting classifiers, Logistic Regression, Naive Bayes, and SVM.
• Retrieval-Augmented Generation (RAG): Dynamic retrieval pipelines with reasoning generation.
• Pathway Framework: Advanced semantic search using Vector Store.
• Google Drive API: Integration for live document updates.
• Multithreading: Ensured efficient operations without downtime.

---

🏆 Hackathon Details

Kharagpur Data Science Hackathon 2025

• Organized by: Kharagpur Data Analytics Group,IIT Kharagpur
• Participants: 🌍 10,000+ participants across multiple rounds.
• Achievement: Proudly emerged as one of the top 10 teams in the final round after rigorous evaluations. 🎉

---

⚙️ How It Works

1. 📤 Upload a Research Paper:
   Use the Streamlit-based interface to upload your research paper in PDF format.

2. 🔄 Preprocessing and Publishability Check:
   The system preprocesses the uploaded paper and predicts publishability using trained machine learning classifiers.

3. 🔗 Semantic Retrieval:
   Relevant reference papers are retrieved using Pathway’s Vector Store based on semantic similarity.

4. 📊 Classification and Rationale:
   The paper is classified into one of the target conferences (e.g., CVPR, EMNLP, NeurIPS, etc.) using RAG. The rationale is generated using Gemini AI for better transparency.

5. ✅ Results Display:
   The Streamlit UI displays the classification and reasoning in real-time for user review.

---

🚀 Installation and Setup

1. 📦 Clone the Repository:

   git clone https://github.com/Neural-Knight/ResearchPaper-ConferenceClassifier.git
   cd WebApp

2. 🔧 Install Dependencies:

   pip install -r requirements.txt

3. 🔗 Set Up Pathway Vector Store:

   • Ensure Pathway is installed and configured.
   • Update the Google Drive credentials file (credentials.json).

4. ▶️ Run the Application:

   streamlit run app.py

5. 🌐 Access the UI: Open your browser and go to http://localhost:8501 to use the application.

---

📂 Folder Structure

The repository is organized into two main directories for clarity and separation of concerns:

1. 🧪 Task

• Contains the Jupyter Notebook (KDSH_Task_Final.ipynb) outlining the steps, experiments, and analysis for the competition tasks.

Files and Folders:

• 📓 KDSH_Task_Final.ipynb: Notebook detailing the problem-solving approaches and methodologies for Task 1 and Task 2.
• 📁 Reference/: Folder containing reference research papers used for initial testing and development.
• 📄 Papers/: Folder containing different type of reasearch papers for testing the model
• 🗂️ Samples/: Folder containing sample reasearch papers for testing the model

---

2. 🌐 WebApp

• Houses the Streamlit-based application and core logic for research paper classification and RAG-based conference prediction.

Files and Folders:

• 🖥️ app.py: Main Streamlit application file to run the UI.
• ⚙️ model.py: Code for machine learning models and preprocessing logic.
• 📁 Reference/: Folder containing dynamically updated reference research papers for Pathway’s Vector Store.
• 📦 requirements.txt: Python dependencies for the project.
• 📤 uploaded_files/: Stored the uploaded papers in the webapp locally.

---

Other Files

• 📝 README.md: Comprehensive documentation of the project, including setup instructions and workflow.

---

🌱 Future Improvements

• 🧮 Integration of Hybrid Index (BM25 + Vector Search) for enhanced retrieval performance.
• 📈 Improved scalability for larger datasets.
• 🤖 Enhanced reasoning capabilities with more advanced language models.

---

💡 Creators

This project was developed by:-

• Malyadip Pal
  
  Undergraduate Student, Department of Electrical Engineering, IIT Kharagpur

• Shaunak Majumdar
  
  Undergraduate Student, Department of Mechanical Engineering, IIT Kharagpur

❤️ Acknowledgments

We extend our gratitude to Kharagpur Data Analytics Group and Pathway, the organizers of the Kharagpur Data Science Hackathon 2025 for providing an excellent platform to showcase and enhance our technical skills.