This repository contains code and data related to the paper "Learning Ideological Embeddings From Information Cascades" by Corrado Monti, Giuseppe Manco, Çigdem Aslay and Francesco Bonchi, published at CIKM 2021. If you use the provided data or code, we would appreciate a citation to the paper:
@inproceedings{monti2020learningideological,
title={Learning Ideological Embeddings From Information Cascades},
author={Monti, Corrado and Manco, Giuseppe and Aslay, Çigdem and Bonchi, Francesco},
booktitle={Proceedings of the 30th ACM International Conference on Information and Knowledge Management (CIKM)},
year={2021}
}
Here you will find (i) code to apply the algorithm we designed on your data (ii) the Reddit and the (anonymized) Twitter dataset we used in the paper (iii) code to reproduce our experiments.
In ideology2vec.py, we provide a Python module to use our algorithm on your data. It requires tqdm and Tensorflow.
This module contains a function called compute_ideological_embeddings that turns propagations of items with known topics on a graph into multidimensional polarities for the nodes of the graph. To understand the details, consult the documentation of the function or the ideology2vec-example.ipynb notebook, that contains a toy example for its usage.
In data/, we provide the data sets we used in our paper:
-
reddit-politicscontains data about which subreddits posted certain URLs. -
twitter-vaccinescontains an anonymized retweet data set: no text, usernames, or any other personal info is provided, only a bipartite graph of retweets.
For both data set, there is a README in their directory providing acknowledgements and further details.
In order to reproduce our experiments, we provide our scripts in src/. They need a set of dependencies, listed in environment.yml. To install them and run experiments, do:
conda create --name multid --file environment.yml
conda activate multid
cd src
python synthetic.py
python main.py twitter
python main.py reddit