New TT-SR Framework Unlocks Sender-Receiver Communities in Directed Networks

Duy Hieu Do's new paper proposes TT-SR (Two-Tier Sender-Receiver), a community detection framework for directed networks that bridges the gap between vertex-level and edge-level clustering. Traditional methods assign a single label to each node, losing the asymmetry of directed edges, while full edge clustering is hard to interpret. TT-SR assigns each vertex a sender role and a receiver role, then assigns each directed edge a type based on the sender role of its source and the receiver role of its target. This yields communities that capture both source-target relations and role constraints. The framework generates candidate assignments from four spectral views: count-residual, stationary-flow, degree-corrected, and order-score, then refines them with local role updates. The final selection uses a two-tier criterion: a degree-corrected profile score as the primary structural signal, with Bernoulli density and order-flow scores as secondary ties. The author justifies the spectral views via sender-receiver modularity relaxations and interprets the degree-corrected score as a likelihood-based residual comparison.

Experiments on three synthetic benchmark types (pathway-type, co-block, ordered-flow) across three scale settings show TT-SR achieves the strongest or essentially tied strongest edge-community recovery, with gains most pronounced on degree-corrected co-block and ordered-flow graphs. On real networks, TT-SR aligns well with Email-Eu-core metadata (a communication network) and extracts strong sender-receiver bicommunity summaries from unlabeled directed networks. The paper is 25 pages, including mathematical foundations, algorithm details, and comprehensive evaluation. This work advances directed community detection by offering a more expressive yet interpretable method than existing approaches, with potential applications in social networks, citation networks, and biological interaction networks where directionality matters.