ParliaMint Topic Detection & Summarization

This project is a prototype aimed at exploring topic detection methods and summarization models for the development of a streamlit dashboard. The idea of the dashboard is to

1. Topic Detection:
   Visualize the prevalence of specific topics in parliamentary debates over time in order.
   • this can be used for a long-term analysis of parliamentary debates or
   • to identify specific debates relevant to the user
2. Summarization:
   For a selected date (e.g. the user has identified a certain day, where a certain topic was debated) the dashboard features a mini-summary for each debate that took place on that day.

Disclaimer: This project is a prototype developed within 12 hours and is under active development. The current data base is the 2022 subset of the ParlaMint parliamentary data of the Netherlands. The current topic detection and generative text summarization models can be seen as placeholder models for demonstrative purposes only. See also the discussion below on further ideas and issues.

How to use

As this is a prototype, the usage instructions are subject to change based on ongoing evaluations and experiments.

Online Dashboard

To interact with the prototype, we have developed a Streamlit application that provides a user-friendly interface for topic detection and summarization.

You can access the deployed version of the Streamlit application through the following link:
https://parliamint.streamlit.app/

Local Installation

1. Clone the repository:

   git clone https://github.com/lutz-he/parlia-mint.git
   cd parlia-mint

2. Install the required dependencies:

   pip install -r requirements.txt

To start the Streamlit application, run the following command in your terminal:

streamlit run streamlit_app.py

Once the app is running, you can access it in your web browser at http://localhost:8501.

Current Issues & Ideas

data & models

• Streamline application to other subsets of the data
• Find and make work a zero-shot classifier for topic modelling on candidate topics
• Evaluate and search for alternatives for summarization model

streamlit app

• Enable clickable events on the graph to subset date selection for debate summaries
• Fix workaround "sneaky_bars" & "sneaky_points" (failed to dodge bars and points, so I tweaked the x-axis values depending on the grouping variable)
• Work on design
• Work on more intuitive UX

Contributing

Contributions are welcome! If you have ideas for improving the topic detection methods or summarization models, please feel free to submit a pull request or open an issue.

License

This project is licensed under the MIT License. See the LICENSE file for more details.

Acknowledgements

Component	Resource
Data	ParlaMint parliamentary data of the Netherlands
Topic detection zero-shot classifier	valhalla/distilbart-mnli-12-3
Summarization	t5-small