This project is developed as part of the following research paper:

F. Gullo, D. Mandaglio, A. Tagarelli. "Polarized Communities meet Densest Subgraph: Efficient and Effective Polarization Detection in Signed Networks" published in ACM Transactions on Knowledge Discovery from Data (TKDD), 2025

This research codebase includes the implementations of the proposed greedy2PC and greedy2PC++ algorithms for discovering polarized communities in signed networks, together with existing baselines (eigensign variants, bansal, local search, etc.). Each dataset is a signed edge list (see datasets/), and every algorithm implementation is in code/algorithms/. The code/main.py entrypoint glues everything together: it parses the CLI, loads the requested signed graph, runs the chosen method, prints intermediate information, and stores structured results under output/<dataset>/<algorithm>_results.json.

• numpy==2.3.5
• scipy==1.16.3

The code was tested with Python 3.11.14.

From the code/ directory, run the following command:

python main.py <dataset> <algorithm> \
  [-p/--print_results] [-b B] [-lsmi LS] [-ct CT] \
  [-T ITER] [-uc/--use_convergence]

• dataset: any filename in datasets/ without the .txt suffix (e.g. congress, slashdot, wikiconflict…).
• algorithm: greedy2PC or greedy2PC++ (proposed methods). The CLI also accepts the paper’s baselines eigensign, eigensign-binary, random_eigensign, bansal, random_local, and greedy for comparison.
• -T / --use_convergence: control the refinement iterations and stopping rule for greedy2PC++.
• -b: multiplicative factor (float or l1) for random_eigensign.
• -lsmi / -ct: maximum iterations and convergence threshold for random_local.

Examples:

python main.py congress greedy2PC

python main.py congress greedy2PC++ -T 10 --use_convergence

Every execution writes (or updates) a JSON entry like the existing output/congress/greedy2PC_results.json. The file contains:

• solutions: the two polarized communities (S1, S2) derived from the final vector solution.
• polarity_scores and agreement_ratios: scalar metrics per solution, as computed in commons.py.
• running_time: wall-clock seconds.
• Algorithm-specific extras (beta, maximum_inconsistent_degree, iterations, etc.).
• parameters: the subset of CLI arguments that influenced the run.
• timestamp: when the experiment finished.

Delete or rename the output file if you want to keep multiple experiment histories; otherwise the script transparently updates the entry matching the same parameter tuple.