We approached this hackathon project by considering how the indico.io API could be used to compare opposing viewpoints on various divisive topics. Ultimately, we decided to use indico.io’s “Custom Collections” feature, which allowed us to train a model on for/against arguments from procon.org, as well as various other sources. This let us input text to predict whether it was for or against a given argument.
Tech stack:
- Python
- Flask
- Indico.io API
- HTML/CSS
- Sass, jQuery
Developers:
- Clone the repo
- Add a
config.pyfile to the project directory and set a secret key for CSRF as follows:
app.config["SECRET_KEY"] = <your key>
- Get an indico.io API key and add it to your home directory in an .indicorc file as explained in the indico.io docs.
- Install dependencies with
pip install -r requirements.txt - cd into the project repo and run
python app.py - In the browser, open http://localhost:5000/predict
- Select topic from dropdown menu
- Enter text passage to predict in input, then select "Submit" button
- Once page has finished reloading, open "Analyze" tab to view prediction
POST /collect/text
Request body format:
{
"text": string,
"label": string,
"tag": string
}
Text: passage of text to train model on
Label: name of model to train, eg: gun_control, climate_change
Tag: pro or con