PerspectroScope

A Django-based web demo system for discovering diverse perspectives to claims.

Here is a video summarizing the main idea:

Setting up the system

Getting started

To install the python dependencies, simply do:

pip3.6 install -r requirements.txt

Check if Django is installed:

$ python -m django --version

Setup the database

> python3 manage.py makemigrations app  # you should see the migrations under "app/migrations"
> python3 manage.py sqlmigrate app 0001
> python3 manage.py migrate

Run the app:

$ python3 manage.py runserver

Details

The system has two layers:

Information Retrieval

The system using a retrieval engine in order to speed up the system. To begin with, download an instance of a retrieval engine from https://elastic.co . And index the two pools of perspective and evidence (included in the perspectrum repository). The two indices have to be named "perspective_pool_v0.2" and "evidence_pool_v0.2", respectively. Here is a brief instruction on how to use elasticsearch and how to index json files.

Google Analysis are done using the following app: https://pypi.org/project/django-google-analytics-app/

Learned models

Download the trained model files from this link and put them under data/model/ (for instance, data/model/relevance/perspectrum_relevance_lr2e-05_bs32_epoch-0.pth).

Citation

If you find this useful, please consider citing the following work(s):

@inproceedings{chen2019perspectroscope,
  title={PerspectroScope: A Window to the World of Diverse Perspectives},
  author={Chen, Sihao and Khashabi, Daniel and Callison-Burch, Chris and Roth, Dan},
  book={ACL - Demos},
  year={2019}
}