# `Text.WordCloud.Backends.YAKE`
[🔗](https://github.com/kipcole9/text/blob/v0.6.2/lib/word_cloud/backends/yake.ex#L1)

YAKE! (Yet Another Keyword Extractor) backend for `Text.WordCloud`.

Implements the unsupervised, statistical keyword-extraction algorithm
described in [Campos et al., *Information Sciences* 509,
2020](https://doi.org/10.1016/j.ins.2019.09.013). YAKE! computes five
per-word features (casing, position, frequency, relatedness to
context, sentence dispersion) entirely from the input document, then
composes them into n-gram candidate scores. No reference corpus or
trained model is required — this is what makes it the right default
for a multilingual word-cloud library.

The algorithm's only language-specific dependency is the stopword
list, supplied via `Text.Stopwords.for/1` (or the caller's `:stopwords`
override). YAKE!'s own design treats stopwords as phrase-boundary
markers and as low-content interior fillers, so a good list directly
improves output quality.

### Score direction

YAKE!'s published score is "lower = more important". This module
inverts internally before returning, so the value passed to the
orchestrator is the standard "higher = more important" form every
other backend uses.

### Options

* `:ngram_range` — `{min, max}` candidate length. Defaults to
  `{1, 3}` (the YAKE paper's default).

* `:window_size` — neighbour-context window for the relatedness
  feature. Defaults to `1` (immediate neighbours), matching the
  reference implementation.

Standard `Text.WordCloud` orchestrator options (`:language`,
`:stopwords`, `:case_fold`, `:locale`) are honoured.

### Caveats

This is a faithful but simplified port of the algorithm: the five
features are computed exactly as in the paper, but the
candidate-generation rules use the stricter "phrases must start
and end with a non-stopword" form rather than the paper's full
composition rules. In practice this produces output well-correlated
with the reference Python implementation (`LIAAD/yake`); a
differential-fixture test against that implementation is a
follow-up.

---

*Consult [api-reference.md](api-reference.md) for complete listing*