BLACKBONE™ Research

Geopolitical Discourse Graph

From NLP classification to source-aware discourse exploration. This research began as a geopolitical NLP project and expanded into a graph-based method for comparing U.S. discourse with affected-side, regional, and non-U.S. perspectives.

Open interactive graph Open research brief Launch presentation View original NLP project

Research claim

The central finding is that source perspective must be governed as a methodological layer. Source retrieval can drift toward dominant archives unless source-origin boundaries, citation checks, and human review are built into the workflow.

528

Evidence passages

Balanced project corpus organized by event, analytical role, and source-perspective layer.

0.9525

Source-layer F1

Source perspective remained the strongest and most separable modeling signal.

4,816

Sentence frames

Sentence-level frame extraction supports actor/action/target graph cards.

Charts

Modeling and graph-layer results

Course baseline macro F1 chart Macro F1 comparison chart Graph layer counts chart

Public boundary

The public research surface may include methodology notes, selected charts, cited sentence-level excerpts, and a future public-mode graph. It does not publish the full raw corpus, full passage text dataset, source collection folders, raw sentence audits, or development graph payloads.