clean & unify tags for packages

[iamYannC]

ggplot2 gallery: tags summary

2026-04-30

Following a misuse of tags myself, as a new ggplot2 extender, I decided
to take a deeper dive into tags in the ggplot2 extension gallery.
This full Quarto doc can be found here.

_config.yml lists every extension with a free-text tags field. Because the field is unconstrained, the field nearly useless for meaningful discovery.

TL;DR

• 158 packages in the extension gallery.
• 193 unique tags after lower-casing (one case-only collision was
  Visualization vs visualization).
• On average, each package has 3.15 tags (md = 3, sd = 1.58).
  ggDNAvis leads with 12 unique tags.
• visualization is the number one tag with 145 appearances;
  general follows with 62. Together they account for 42% of all
  tag uses but carry essentially no information.
• The tag distribution is extremely long-tailed: 148 of 193 tags (77%)
  are used by exactly one package, and only 20 tags reach n ≥ 3.

What is wrong with the current tags

The raw .qmd file contains an object tags_in_pkgs with the summary
information. download & explore it for better understanding.

1. Case / spelling / number variants that should clearly merge:
   Visualization vs visualization vs visualisation vs
   visualizations; geom vs geoms; time series vs time-series;
   theme vs themes; facet vs facets; outlier vs outliers;
   customisable vs customizable; algorithm vs algorithms.
2. Filler tags that carry no signal. general is applied by 62
   packages and tells a reader nothing. visualization is applied by
   145 packages — i.e., almost everyone — so it is not a discriminator
   either. I mean, this IS a ggplot2 gallery after all.
3. Specialized niche vocabulary. Several packages invent their own
   private taxonomy. Examples:

• gganatogram: anatograms, tissue, anatomy, expression, pharmacology
• ggDNAvis: DNA, RNA, customisable, customizable, medicine, methylation, sequence, FASTQ, ...
  (12 tags, all unique to this package).
• ggblend: blending, affine transformation, layer algebra, compositing` (4 tags, all unique).

These are accurate descriptions but they cannot help anyone find the package, because no other entry uses the same words.

4. Redundant within a single entry. ggoutlierscatterplot lists
   both outlier and outliers, both algorithm and algorithms.
   ggDNAvis lists both customisable and customizable.

Proposed unification

The core strategy is to reduce and unify tags to clusters that provide
actual information:

• Normalise all tags to lower-case and trim whitespace.
• Collapse spelling / case / number variants to a single canonical form.
• Group small related tags (n < 3) under a topical umbrella when a
  clear cluster exists.

Better control in the future

1. Drop or rename general and visualization. Both are applied
   broadly enough that they fail to discriminate between packages and
   add little to no signal.
2. Encourage informative, topic-based tagging. Publish a suggested
   vocabulary in the contributing guide, organized by topic
   (life-sciences, spatial, time-series, distributions, etc..). Help us
   help the potential future user
3. Tidymodels has a reactive table. Consider using one for the gallery too?

Reproducing the analysis

This analysis relies on _config.yml
from commit ed45cf8 and is fully parameterized in the script.
Download the full script here: ggplot2-gallery-tags-summary.qmd

Yann

[iamYannC]

i know that's a lot but it's a one-time cleanup and a small adition to contributor guide (suggested tags) - from there things should be way easier. the end goal is to allow users (that is, people who use ggplot2) to better search and find what they need, in the topic they're looking..