Skip to content

Visualization

Jump to bottom
Bell Eapen edited this page Nov 10, 2025 · 1 revision

Visualization

Overview

Visualization within CRISP-T (via crispviz) transforms textual and numeric analyses into interpretable graphical artifacts that facilitate sense-making and triangulation. Visual outputs help researchers align emergent qualitative themes (topics, sentiments) with quantitative structures (clusters, PCA components, TDABM graphs) to refine theoretical categories.

Implemented Visualizations

Visualization Command Purpose
Topic Model Charts crispviz --ldavis Explore topic-term distributions for thematic comparison
Word Cloud crispviz --wordcloud Surface dominant lexical elements; corroborate coding emphasis
PCA Scatter crispviz --pca Show numeric clusters and variance structure for category mapping
TDABM Graph crispviz --tdabm Reveal topological connectivity supporting structural memos
Distributions / Histograms crispviz --dist (bins configurable) Inspect feature distributions; identify skew linked to narrative patterns

Sense-making Integration

  • Use topic visualizations to validate whether high-importance regression or decision tree features correspond to distinctive textual themes.
  • Employ PCA scatter plots to locate outlier points; retrieve associated documents and compare their narrative content for theoretical refinement.
  • Overlay cluster labels (K-Means) conceptually with topic prevalence: convergence suggests robust multi-modal categories.
  • TDABM graphs can be juxtaposed with word clouds of documents mapped to specific balls for localized thematic interpretation.

Example (Business Domain)

A PCA plot reveals a discrete cluster of customers with high engagement metrics. The word cloud of associated feedback emphasizes "workflow automation". Topic visualization shows a dominant topic referencing integration. These converging visuals support a grounded category on efficiency-oriented adoption.

Example (Medical Domain)

TDABM graph colored by fatigue score highlights a dense region. Word cloud of documents in that region shows terms "sleep," "pain," "energy." PCA loadings indicate sleep_balance and inflammation_marker contribute strongly. Visual triangulation yields a refined theoretical construct about inflammatory-sleep interplay.

Workflow Recommendations

  1. Generate analytical metadata first (--topics, --assign, --ml).
  2. Produce visualizations with crispviz referencing the same --inp corpus.
  3. Annotate visuals in research memos, linking observed graphical patterns to qualitative codes.
  4. Use semantic search on visually identified subsets to extract exemplar narratives.

CLI Usage

crispviz --inp crisp_input --wordcloud --out viz_out/
crispviz --inp crisp_input --ldavis --out viz_out/
crispviz --inp crisp_input --pca --out viz_out/
crispviz --inp crisp_input --tdabm --out viz_out/

Best Practices

  • Centralize output in a version-controlled viz_out folder.
  • Record radius values and topic counts directly alongside images for reproducibility.
  • Combine quantitative and qualitative figure captions: e.g., "Cluster 2 PCA region; narratives emphasize workload, numeric features show high travel_time."

Future Enhancements

  • Interactive dashboards combining TDABM and topic association.
  • Automatic generation of memo templates for each visualization.
  • Heatmaps of topic prevalence across PCA or Ball Mapper coordinates.

References

  • Mettler et al. (2025) — Reflective methodological guidance.
  • Rudkin & Dlotko (2024) — Structural exploration via Ball Mapper.

See Also

Clone this wiki locally