When investigating organizational behavior—especially within large public institutions or regulatory bodies—traditional document search often fails to answer macroscopic questions: Who is coordinating on what topic? When did internal policy focus shift? Which individuals serve as information bottlenecks or bridges?
Graph-based communication analysis provides a mathematical lens to unpack these structures.
The Analytical Framework
Analyzing thousands of unstructured institutional communications (emails, memos, and logs obtained via Public Records Act requests) requires a disciplined pipeline:
1. Entity Resolution & Thread Reconstruction
Raw email data is notoriously noisy: individuals use personal aliases, department aliases, and fragmented forwarding chains. Canonical entity resolution maps diverse email headers into persistent actor nodes, while subject heuristics and In-Reply-To headers reconstruct conversational subgraphs.
2. Temporal Windowing (Time-Slicing)
Static graphs hide organizational dynamics. By segmenting message networks into weekly or monthly rolling windows with exponential decay on edge weights, we can observe:
- Sudden spikes in topic intensity.
- Structural reconfigurations following external catalysts (such as court filings, public audits, or leadership changes).
Measuring Influence & Community Structure
- Weighted Degree Centrality: Identifies the primary volume hubs within departments.
- Betweenness Centrality: Surfaces critical “broker” nodes—individuals who connect otherwise isolated operational silos.
- Modularity Clustering: Algorithms like Louvain or Label Propagation group actors into functional communities based purely on interaction topology, highlighting informal working groups that may differ from official organizational charts.
Lightweight Topic Coupling
Pairing graph topology with temporal TF-IDF or Latent Dirichlet Allocation (LDA) allows us to tag each cluster and time window with distinct thematic keywords. Rather than reading thousands of mundane operational notes, investigators can immediately pinpoint the specific nodes and dates corresponding to topics of legal interest.