Text2Fillups

Text2Fillups is a web-based ML application that allows users to generate fill-in-the-blank questions from given texts. This repository contains the code for both the backend server and the frontend user interface, as well as model training and inference notebooks.

📕 Tech stack

📽️ Video demo

2023-07-09.14-06-19.mp4

🔥 Web-app Features

💻 Stylish glassmorphism UI

• The UI has been created using material-ui react library.

📜 Toggle Feature for questions

• The toggle feature has been implemented in the question answer cards to hide and unhide the answers.

🍞 Toast Messages

• Toast messages are displayed wherever necessary, to indicate a completion of a process, for a better accessibility.

🔗 Google Quiz link

• Google quiz link is readily generated for assessment of the topic.

🧠 Overview

1. The text corpus is taken as input from the user and question-answer pairs are generated using the t-5 transformer model from hugging face for the task of context-aware question generation. (generates question and answer pairs using a context).
2. Then these question-answer pairs are converted to normal sentences (For e.g.: {"Question":"What is your name ?", "Answer":"Dhanesh"} is converted to "My name is Dhanesh") using t-5 transformer fine-tuned on QA2D dataset.
3. The fine-tuned model is pushed in hugging-face hub for further usage.
4. To reduce the inference time and reduce the model size, I have quantized the fine-tuned transformer using ONNX and fast-t5 library that reduced the size of the model from around 900MB to around 400MB. The models and relevant notebooks are uploaded in Fine-tuning-notebook/Quantized-model.
5. The answer word is blanked and displayed in the web app with a toggle feature.
6. Additionally, a Google quiz link containing all these questions is created using google forms API in the backend, which can be shared to test the topic readily.

📊 Metrics from Weights and Biases

I used weights and biases to log metrics like epochs, train_acc, valid_acc, BLEU score etc. Click here to view them.

🙌 Instructions to Get Started

• First of all, git clone this repository and go to the appropriate folder in your local machine.

📜 Getting credentials for GoogleFormsAPI

1. Create an account on https://console.cloud.google.com/ and create a new project.
2. In the API & Services section enable the API for Google Forms.
3. Generate a new OAuth 2.0 Client ID and download the json file containing Client_ID and Client_Secrets.
4. Save this json as credentials.json in the backend folder.

🔌 Backend Server

1. Navigate to the folder where this repository is cloned.
2. Open the terminal and navigate to the backend directory using the command: cd backend.
3. Install the required dependencies by running: npm install.
4. Start the backend server by running: npm start.
5. Install the Python dependencies by running: pip install -r requirements.txt.
6. Run the Flask application by executing: python app.py.

👤 Frontend User Interface

1. Open a new terminal window.
2. Navigate to the frontend directory using the command: cd frontend.
3. Install the necessary dependencies by running: npm install.
4. Start the frontend development server by running: npm start.

That's it! You are now ready to use Text2Fillups and generate fill-in-the-blank questions from texts. Enjoy! 🚀📚

🦾 Contributors

• Dhanesh ,CSE , IIT Guwahati.

🌟 Stay connected

Don't forget to ⭐️ star this repository to show your support and stay connected for future updates!