Explorando os Limites da Reprodutibilidade na Tarefa de Detecção de Comunidades em Modelos de Redes

Context: Network model-based studies are applied across various fields, including social media. Problem: Data availability and reproducibility are challenging due to restrictive data policies and the significant computational resources required to process large and complex networks. In this context, sampling techniques offer a viable alternative by selecting representative sub-networks that preserve the essential structural properties of the original network. Despite their potential, there is a dearth of studies investigating how sampling methods can generate networks at different scales and quantify their limitations in detecting communities, especially in conjunction with backbone extraction, a crucial step that can significantly affect the network’s probabilistic properties. Solution: This paper evaluates the effectiveness of different sampling methods in improving the availability and reproducibility of network analysis, with a focus on community detection and backbone extraction. SI Theory: Our work is supported by Social Network Theory, which emphasizes relationships and connections among actors within a network over individual attributes. Method: In our research we quantify the limits, properties and scenarios in which smaller network versions can provide comparable communities structures to the original network. Summarization of Results: Our results show that certain sampling methods can effectively capture community structures even in reduced network representations. Contributions and Impact in the area of SI: This research facilitates the reproducibility and democratization of network studies and provides guidelines for creating networks at different scales that allow researchers to replicate certain studies.