A benchmarking tool for the generation of bipartite network models with overlapping communities

Many real-world networks display hidden community structures with important potential implications in their dynamics. Many algorithms highly relevant to network analysis have been introduced to unveil community structures. Accurate assessment and comparison of alternative solutions are typically approached by benchmarking the target algorithm(s) on a set of diverse networks that exhibit a broad range of controlled features, ensuring the assessment contemplates multiple representative properties. Tools have been developed to synthesize bipartite networks, but none of the previous solutions address the issue of generating networks with overlapping community structures. This is the motivation for the BNOC tool introduced in this paper. It allows synthesizing bipartite networks that mimic a wide range of features from real-world networks, including overlapping community structures. Multiple parameters ensure flexibility in controlling the scale and topological properties of the networks and embedded communities. BNOC’s applicability is illustrated assessing and comparing two popular overlapping community detection algorithms on bipartite networks, namely HLC and OSLOM. Results reveal interesting features of the algorithms in this scenario and confirm the relevant role played by a suitable benchmarking tool. Finally, to validate our approach, we present results comparing networks synthesized with BNOC with those obtained with an existing benchmarking tool and with already established sets of synthetic networks, in two different scenarios.