Answers are grounded in this paper and cite their sources.
Demonstrates actionable interventions for network nodes, suggesting new pathways for user agency.
Algorithmic decisions on network-structured data, such as community detection, often remain opaque, leaving individuals without agency over how they are categorized. This paper introduces a general framework for algorithmic recourse on networks, designed to provide actionable recommendations for nodes seeking to alter their assigned properties. Shifting the focus from descriptive interpretability “Why?” to prescriptive actionability “How?”, we formulate recourse as a nearest-counterfactual search constrained by an actionable set, which is a subset of edges a user has the power to modify. We introduce a weight-aware notion of responsibility that functions as a feasibility metric, quantifying the minimal cost of interventions required to achieve a desired outcome. To demonstrate the framework's broad applicability, we apply it to the community detection problem, where the goal is to identify the most feasible edge modifications that allow a node to leave a specific community. We provide a theoretical treatise for a modification method that enables an efficient search procedure with provable approximation guarantees. Our experiments on a diverse suite of real-world networks demonstrate that our method achieves high success rates with low-cost interventions in both weighted and unweighted scenarios. Furthermore, comparisons with a state-of-the-art Deep Reinforcement Learning Agent reveal that our interpretable framework matches or exceeds, in most cases, the performance of complex black-box methods. The primary contribution of this work is an interpretable and tenable framework for network recourse, emphasizing that transparent, structural manipulation is a robust alternative to complex learning-based approaches for restoring user agency.
No takes yet. Share an insight, caveat, or question.
Konstantopoulos et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: