Update: Use this colab notebook to reproduce our results

If you use this code or part of it, please cite the original reference:

Minici, M., Cinus, F., Monti, C., Bonchi, F., & Manco, G. (2022, October). Cascade-based echo chamber detection. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management (pp. 1511-1520). doi: https://doi.org/10.1145/3511808.3557253

Before running any experiment, we suggest you install the provided conda environment.

How to install the conda environment: conda env create --file environment.yml

In both sections, the parameters highlighted in bold are the ones you should modify in order to fully reproduce our results.

In order to run a synthetic experiment, you can use python grid-synthetic.py if you want to use default settings, otherwise, you can specify several options:

• seed, to change the initial random seed for stochasticity;
• N, to change the number of initial users;
• eta, to change the polarity of each community;
• s, to change the social prior size;
• h, to change the echo-chamber prior size;
• B, you can ignore this parameter;
• items_per_node, to change the number of items per user;
• lr, learning rate of ECD procedure;
• epochs, number of epochs of ECD procedure;
• oversampling, whether you want to oversample the minority class --- either edges or propagations;
• reweighting, whether you want to reweight the two classes w.r.t. their cardinality;
• annealing, whether you want to add an annealing procedure;
• ablation, whether you want to exclude links or propagations from the ECD procedure;
• training_type, you can ignore this parameter;
• model_type, you can ignore this parameter;
• device, whether you want the experiment to be run on GPU. If no GPU is available, then pass an empty string.

In order to run a synthetic experiment, you can use python grid-real.py if you want to use default settings, otherwise, you can specify several options:

• seed, to change the initial random seed for stochasticity;
• dataset, to change the dataset to analyze. Pick one between brexit and vaxNoVax.
• K, to change the number of communities you want to find;
• s, to change the social prior size;
• h, to change the echo-chamber prior size;
• • B, you can ignore this parameter;
• lr, learning rate of ECD procedure;
• epochs, number of epochs of ECD procedure;
• oversampling, whether you want to oversample the minority class --- either edges or propagations;
• reweighting, whether you want to reweight the two classes w.r.t. their cardinality;
• annealing, whether you want to add an annealing procedure;
• link_removal, you can ignore this parameter!
• prop_removal, whether you want to exclude some propagations from the dataset;
• prop_removal_perc, the percentage of propagations you want to remove;
• stance_detection_exp, it is a boolean flag --- i.e.: True or False --- that when you want to analyze the stance detection capability of ECD if set to True, removes all propagations of the analyzed users.
• ablation, whether you want to exclude links or propagations from the ECD procedure;
• training_type, you can ignore this parameter;
• model_type, you can ignore this parameter;
• device, whether you want the experiment to be run on GPU. If no GPU is available, then pass an empty string.