tiny-mood-generate

Build a tiny-mood mood table from your own content, using real GloVe word embeddings instead of a small hand-written word list.

This is a build-time tool, not a runtime dependency. It reads your blog posts, looks up the words that actually appear in them, and writes a small JSON file. Your app never needs this package or GloVe itself at runtime — only the generated JSON.

What it is

tiny-mood's mood scoring works by averaging known words in a lookup table. The problem with a small hand-written table is coverage: words like "acquisition," "regulatory," or "orchestration" simply aren't in it, so any post using that kind of vocabulary scores as falsely neutral.

tiny-mood-generate fixes this by scanning your real content for its actual vocabulary, looking each word up in a pre-trained GloVe embedding, and calibrating the result against a small set of reference words with known mood values — so the judgment already encoded in tiny-mood's defaults carries over to any word GloVe has seen, not just the words someone happened to type into a hardcoded list.

Install

npm install -D tiny-mood-generate

Quick start

npx tiny-mood-generate ./content ./mood-table.json

./content should be a directory of JSON files with title, description, and optionally sections (each with heading/body) — typically your blog post source files.

On first run, this downloads and caches a GloVe embeddings file (~59 MB, one time, reused across all future projects on your machine). You'll be asked to confirm before anything downloads:

tiny-mood-generate needs a GloVe word embeddings file to build your mood table.

Source: Zenodo (Leipzig University mirror of Stanford's GloVe 6B, 50-dimensional)
Size: ~59 MB, one-time download
Cached at: ~/.cache/tinymood/glove.6B.50d.txt
License: Public Domain Dedication and License v1.0

Proceed with download? (y/n)

Every run after the first reuses the cached file instantly — no repeat downloads, no repeat prompts.

// In your app, using tiny-mood (separate package):
import { getMoodBackground } from 'tiny-mood'
import moodTable from './mood-table.json'

getMoodBackground(text, moodTable, seed)

CLI usage

npx tiny-mood-generate <contentDir> <outputPath> [options]

| Flag | What it does | |---|---| | --glove <path> | Use a GloVe file you already have instead of downloading. Accepts plain-text format (word d1 d2 d3...) or the gensim binary word2vec format — both are detected and handled automatically. | | --yes | Skip the download confirmation prompt (for CI / non-interactive use). |

Examples

# Standard use — downloads and caches GloVe automatically on first run
npx tiny-mood-generate ./blog/content ./blog/mood-table.json

# Use a GloVe file you already have
npx tiny-mood-generate ./content ./mood-table.json --glove ./glove.6B.300d.txt

# Non-interactive, e.g. in a CI pipeline
npx tiny-mood-generate ./content ./mood-table.json --yes

What you'll see

Scanning content in ./content...
Generation stats:
  vocabulary words found in content: 84
  calibration anchors requested: 27
  calibration anchors found in GloVe source: 27
  words successfully scored: 84
  words not found in GloVe source (skipped): 0
Wrote 84 word vectors to ./mood-table.json

Use it with tiny-mood:
  import { getMoodBackground } from 'tiny-mood'
  import moodTable from './mood-table.json'
  getMoodBackground(text, moodTable, seed)

Words not found in the GloVe source are silently skipped, the same way an unknown word is skipped by tiny-mood's default table — no crash, no placeholder value, just absence.

Functions

For use in your own build scripts, instead of the CLI:

generateMoodTable(contentDir, glovePath)

Runs the full scan → lookup → calibrate → score pipeline. Returns { table, stats }.

ensureGloveFile(options?)

Handles the download/cache/format-conversion logic on its own — useful if you want to trigger the GloVe acquisition step separately from generation, or build your own CLI on top of it. Returns the local path to a ready-to-use plain-text GloVe file.

Regenerating as content grows

The table is a static snapshot of your vocabulary at the time you ran the command. New posts with new words won't be scored until you regenerate:

npx tiny-mood-generate ./content ./mood-table.json --yes

This is fast after the first run — the GloVe file is cached, so regenerating only re-runs the scan/lookup/calibrate steps, not the download.

Where this fits

your content (JSON)  ──┐
                        ├──>  tiny-mood-generate  ──>  mood-table.json
GloVe embeddings  ──────┘            (build time)          │
                                                             ▼
                                                  tiny-mood (runtime)
                                                  getMoodBackground(...)

tiny-mood-generate and the GloVe file it uses never ship inside your app's runtime bundle or dependency tree — they're a devDependency only, used to produce one small JSON file that tiny-mood reads at runtime.

Try it live

[StackBlitz demo — link coming soon]

License

MIT. GloVe embeddings are from Stanford's GloVe project, redistributed via Zenodo under the Public Domain Dedication and License v1.0.