Computational Frame Analysis

Correlating
Resonance

Identifying frames of reference in contested policy spaces

A structured methodology for revealing the conceptual configurations that underpin conflicting positions within complex or wicked problems. Apply it to any corpus of policy texts to map what different stakeholders agree on, what divides them, and where common ground exists.

Begin analysis →
4 Analytical phases
5 Required outputs
∞ Corpus types
What it does

From a body of texts to a map of frames

Correlating Resonance takes a corpus of texts expressing different positions on a contested issue and produces a structured account of the frames of reference that make those positions incompatible — and the ground on which they might yet converge.

The methodology combines inductive concept coding, document-term matrix construction, and cluster analysis to reveal three analytically distinct types of concept across the corpus.

Core differentiating
Strongly present within one frame, absent from others. These are the concepts that make positions genuinely incompatible.
Core bridging
Present and important across two or more frames. The common ground on which people who disagree can still collaborate.
Boundary concepts
Frame-neutral or internally inconsistent. These do not clearly belong to any frame and are noted but not interpreted as primary outputs.
The process

Four phases, each building on the last

1
Phase One
Typology Development

Upload 5 randomly selected documents from your corpus. The application extracts candidate concepts through open coding, processes each document in isolation to prevent context accumulation, then consolidates into a unified candidate term list.

  • Upload 5 random corpus documents
  • Sequential isolated concept extraction
  • Consolidation pass across all documents
  • Download candidate term list for review
  • Edit definitions, merge duplicates, add synonyms
5 × PDF or TXT Typology CSV
→
2
Phase Two
Saturation Check

Upload your revised typology and a new batch of documents (5–10% of the corpus). Each document is coded against the fixed typology in an isolated session. Any content not captured by existing concepts is flagged for researcher review. Repeat until the typology holds without new concepts emerging.

  • Upload revised typology CSV
  • Upload new document batch
  • Code each document in isolation
  • Review flagged new concept candidates
  • Update typology and repeat if needed
Typology CSV Document batch Saturated typology
→
3
Phase Three
Full Corpus Coding

Apply the saturated typology to the full corpus. Documents are processed in configurable batches with automatic checkpoint exports. A hybrid text and vision extraction method handles all PDF types. Outputs are assembled into a Document-Term Matrix.

  • Upload final typology CSV
  • Upload full corpus (folder or multi-select)
  • Configure batch size and extraction mode
  • Resume from checkpoint if interrupted
  • Export complete DTM for review
Final typology Full corpus DTM CSV
→
4
Phase Four
Cluster Analysis

Upload the reviewed DTM to complete the analysis. Interactive threshold setting with live heatmap, TTM construction, MCA-based cluster analysis with adjustable cluster count, and LLM-assisted frame naming with researcher validation.

  • Set binarisation threshold interactively
  • Build and download sorted TTM
  • Run MCA cluster analysis
  • Compare sorted vs clustered TTM
  • Name and export frames
Reviewed DTM Frame analysis
→
End-to-end workflow
Phase 1 · Step 1
Initial extraction
5 random documents coded in isolated sessions to generate candidate terms
→ Candidate term list
Phase 1 · Step 2
Researcher review
Merge duplicates, enrich definitions with synonyms, apply scaffolding structure
→ Revised typology CSV
Phase 2 · Iterate
Saturation testing
New document batches tested against fixed typology until no new concepts emerge
→ Saturated typology
Phase 3 · Step 1
Full corpus coding
Every document scored against every concept — one isolated API call per document
→ Raw DTM
Phase 3 · Step 2
Binarisation
Scores converted to binary important / not important using fixed threshold rules
→ Binary DTM
Analysis
Cluster analysis
TTM construction and MCA reveals concept configurations — the frames of reference
→ Frame map
Required outputs
Phase 1–2
Final typology
A flat, non-hierarchical list of concepts with neutral definitions written as coding rules, enriched with synonyms across saturation iterations. Required for full corpus coding and external validation.
Phase 3
Raw Document-Term Matrix
Every document scored against every concept prior to binarisation. The primary data record for repeatability assessment and subsequent identification of representative documents per frame.
Phase 3
Term-Term Matrix and thresholds
Co-occurrence counts across the binarised DTM, together with the threshold values used to produce it. Necessary for replication.
Analysis
Cluster analysis and frame map
Identified clusters with constituent concepts, frame strength measures, concept designations (bridging / differentiating / boundary), and the documents most closely matching each frame.
Analysis
Named frames
Working names proposed by the LLM and validated by the researcher against representative documents. May be revised following that review.