Demo website for for CS3244 Machine Learning Project (AY20/21 Semester 2).
The project was to perform analysis of Singpore's parliamentary hansard using NLP. We trained and ran Transformer models on the Singapore Parliament Hansard to perform sentiment analysis, name-entity recognition and summarisation, taking advantage of transfer learning via pre-trained models.
The models were used to analyse the Hansard of recent years to uncover interesting findings on speakers and entities by sentiment. The findings can be found on the demo website.
The project report can be found here.
The data for training and analysed was scraped from the Singapore Parliament website for all sessions from September 2012 to March 2021, giving about a decade worth of information and spanning three sessions of Parliament. The data can be found here.
The trained models along with the code and results for model training can be found here.
This notebook can be used to get a quick overview of the NLP tasks using Hugging Face Transformers library:
- The models for sentiment analysis and name-entity recognition were deployed on Google cloud run using FastAPI and Docker. Deployment code and details can be found here.
- The demo website was deployed using Streamlit. Instructions for setting up the website locally are given below.
- Clone the repo and navigate to the correct folder
git clone https://github.com/nus-cs3244-ml-singapore-7/NER-demo.git
cd NER-demo
- Create a virtual environment
pip install virtualenv #Run this if you don't have virtualenv installed
virtualenv env
- Activate the virtual environment
env\Scripts\activate #Windows
source env/lib/activate #Mac/Linux
- Install the project requirements
pip install -r requirements.txt
- Run
streamlit run app.py - Visit
localhost:8501to view the app