- Aayush Shah - ajshah_b19@it.vjti.ac.in
- Abhishek Joshi - abjoshi_b19@ce.vjti.ac.in
- Dhruvi Doshi - drdoshi_b19@ce.vjti.ac.in
- Kushal Shah - kashah_b19@ce.vjti.ac.in
- Mann Doshi - madoshi_b19@it.vjti.ac.in
- Meet Parekh - mmparekh_b19@it.vjti.ac.in
- User can search for which relevant articles would be fetched.
- User can select which articles he/she wants to summarize.
- Abstract summaries of each unique article is generated. Uniqueness of an article is determined by its cosine distance with other articles.
- User can also upload articles in .txt/.docx format to get their summaries.
- GitHub Repo link: Link to repository
Tools and technologies that were learnt and used in the project.
- Python
- HTML
- CSS
- JavaScript
- Beautiful Soup
- Flask
- PyTorch
- transformers
Clone GitHub Repository
git clone https://github.com/Kushal-Ajay-Shah/In_A_Nutshell.git
Create a venv/conda environment
- For venv:
python3 -m venv /path/to/new/virtual/environment
- For conda:
conda create --name /name/of/environment
Activate virtual environmenrs
- For venv
source /path/to/new/virtual/environment/bin/activate
- For conda
conda activate /name/of/environment
Install requirements
Navigate to the repo directory then:-
- For venv
pip3 install -r requirements.txt
- For conda
conda install --file requirements.txt
- For T5 model
Download following files: Link
- config.json
- pytorch_model.bin
- spiece.model
- tokenizer.json
Make a directory named 't5-base' inside 'model' Place the downloaded files inside 't5-base' directory
- For PEGASUS
Download following files: Link
- config.json
- pytorch_model.bin
- spiece.model
- tokenizer.json
Make a directory named 'pegasus' inside 'model' Place the downloaded files inside 'pegasus' directory
Run the app.py file
python3 app.py
You can now see the demo of the website on http://127.0.0.1:8000/
Sometimes it is required to refer to online news articles to obtain further information about a topic, trend or event beyond what is already known. It would take hours of manual effort to go through these news articles, understand them and assimilate key findings. Often information would be spread out and not available in a single article.
Thus reading a summary might help the user to avoid reading the entire article or may help him decide if he wants to read that article in detail.
Currently, the project provides Single-Doc Abstractive Summary.
Our next goal would be to provide a Multi-Doc Abstractive Summary. Through this the user can get summary of multiple docs thus covering more information while taking into similarity between articles.
