Paper · June 2026
What Heuristic Methods Can Effectively Detect Fragmented Networks in IRS and FEC Data?
An Elicit systematic review of entity-resolution and graph-clustering methods for campaign-finance and nonprofit-filing data.
Elicit Systematic Review (Automated literature synthesis (Semantic Scholar + OpenAlex corpus))
Abstract
Hierarchical fuzzy spectral clustering, modularity-based graph partitioning, Bayesian network inference, graph pattern matching, and federated graph neural networks can all effectively detect fragmented networks in IRS and FEC data. Spectral and modularity methods are best suited for campaign finance community recovery; graph pattern matching is most efficient for structured tax fraud detection; federated and ensemble architectures offer the strongest scalability for large-scale financial networks. Systematic review: 494 papers screened via Elicit (Semantic Scholar + OpenAlex), 25 included for extraction.
Summary
This is an automated systematic literature review produced with Elicit, screening 494 papers (Semantic Scholar and OpenAlex) down to 25 included studies. It underpins the entity-resolution and network-clustering methodology used elsewhere in this research corpus to detect fragmented or duplicated entities across IRS 990 filings and FEC committee/candidate data.
Methodology
Elicit's systematic review workflow: query formulation, automated screening against Semantic Scholar and OpenAlex indexes (494 papers screened), relevance filtering down to 25 included papers, and structured extraction of method, applicability, and scalability claims from each.
Data Provenance & Ingestion Integrity