RTSum

RTSUM: Relation Triple-based Interpretable Summarization with Multi-level Salience Visualization

• Relation combiner model
• Training dataset

It decompose sentences into triple and decomposing sentences into smaller units and recomposing them with the most important information. We integrated the concept of Knowledge Graph and Relation triple into summarization AI by combining extractive summarization and abstractive summarization.

Get Started

This project are using rye (recommended)

rye sync
python -m spacy download en_core_web_sm

Source Code Structure

web.py is for web gui, main.py is for cli.

• web.py use main.py
• main.py use src/summary.py
• src/summary.py use src/extract.py, src/rank.py and src/abstract.py

.env Example

Running OpenIE server is needed for RTSum to work. You should make a .env file in the root OpenIE server requires around 10GB of RAM to run.

OPENIE_URL='http://localhost:8000'

Run CLI

docker-compose up --scale openie5=4 -d
pm2 start web/service.py --interpreter python3
python main.py file 'data/cnn/article.txt'
python main.py text 'I made arrangements pick up her dog'

Run GUI

docker compose up
streamlit run web/simple.py

GUI Demo

AI summarization algorithm which can highlight the important part of the article with inline visualization using highlighting. Red is for relation triple, blue is for sentence, and green is for phrase word.

Develop Experience

• Formatter - Autopep8
• Typing - Mypy
• Linter - Pylint (recommended)