Cluster Ambiguity in Networks as Substantive Knowledge

Visual network analysis has become increasingly used by scholars in the social sciences and humanities (SSH). While ambiguity is an inherent characteristic of this methodology, most currently available tools lack strategies to make such ambiguity transparent, particularly in relation to non-deterministic algorithmic results. The Louvain modularity algorithm, commonly used to detect communities within networks, is a good example of the issue: repeated executions can result in different community assignments for the same nodes. This paper introduces a novel technique for the visual inspection of results produced by the Louvain modularity algorithm. The proposed method involves an edge-centric analysis that evaluates how consistently pairs of connected nodes are assigned to the same community. This consistency metric is then visualised through a dedicated technique that uses colour-coding to highlight both stable and ambiguous relationships between nodes and clusters. The paper demonstrates the effectiveness of this approach with a proof-of-concept applied to benchmark datasets frequently used in the evaluation of network analysis tools. Finally, the contribution reflects on how this visual technique can support and enhance the heuristic practices of SSH scholars.