This project explores equivalence-based role mining using three types of equivalence: structural, automorphic, and regular. It applies algorithms for each equivalence type, including Euclidean distance, SimRank, RoleSim, REGE, and Blockmodelling, to identify and categorise user roles within subgraphs derived from the Der Standard online forum networks capturing the users' behaviour. In case pairwise similarity is produced by an approximation algorithm, further clustering analysis is performed to assign nodes to clusters. Results indicate that the REGE and Blockmodelling algorithms allow us to infer user roles (network positions) and their relationships.
Data is not shared due to agreement with the university
exploration.ipynb,extract_subgraph.ipynb: data inspection and subgraph construction.rolesim.ipynb,simrank.ipynb,role_interpretation.ipynb: Python pipelines for automorphic/regular analyses.blockmodelling.ipynb,clustering.ipynb,rege.ipynb: R notebooks for equivalence diagnostics and role interpretation.graphs/: graphml/edgelist inputs and label mappings (data sharing restricted externally).docs/: intermediate/final presentations detailing results and limitations.- image files in the
results/folder are mainly for the presentation and final report.
For the R code, after installing the required libraries, it is possible to simply run each notebook from top to bottom. Code cells that take a long time to run are skipped by default. This behaviour can be configured by simply setting the corresponding switch variable.
For the Python code, first install the requirements in the .txt file of the respective .py files. E.g., requirements_rolesim.txt for rolesim.ipynb.
Doran, D. (2017). Equivalence-Based Role Mining. In: Network Role Mining and Analysis. SpringerBriefs in Complexity. Springer, Cham. https://doi.org/10.1007/978-3-319-53886-0_3