Models

A collection of simple data models for text processing in JavaScript. Zero dependencies between modules — import only what you need.

Modules

Trie

A compressed prefix tree (radix tree) for fast word lookup, autocomplete, and fuzzy matching via Levenshtein distance.

       ┏━━━━┻━━━━┓
       P         CA
       ┃         ┃
       ┃         ┣━━ R     CAR
     ┏━┻━┓       ┗━━ T     CAT
     A   OT                POT
     ┃
     ┣━━ RT                PART
     ┗━━ S ━━ S            PASS
              T            PAST

Single-child nodes are automatically merged during construction (e.g. C + A → CA), reducing tree depth and lookup time.

import Trie from './trie.js';

const trie = new Trie({ input: ['car', 'cat', 'cart', 'card', 'pot'] });

// Autocomplete — returns the first match by default
trie.complete('ca');        // 'car'
trie.complete('ca', 5);    // ['car', 'cat', 'cart', 'card']

// Exact lookup
trie.contains('cat');       // true
trie.contains('cab');       // false

// Fuzzy search — find words within a Levenshtein distance
trie.closest('crt', 2);    // [{ word: 'car', distance: 1, index: 1 }, ...]

API

Method	Returns	Description
new Trie({ input })	Trie	Build from a string or string[].
complete(prefix, maxWord?)	string | string[]	Autocomplete. Returns a single string when maxWord is 1 (default), an array otherwise.
contains(word)	boolean	Exact membership test.
closest(target, maxDistance?)	Array<{ word, distance, index }>	Fuzzy matches sorted by edit distance, with a frequency-based index penalty for less common words.
findWords(prefix, maxWord?)	string[]	All words under a given node (internal, but public).

---

N-gram

A statistical language model that predicts the next word based on the preceding n − 1 words. Useful for autocomplete and text suggestions.

                  [1.00]: THE
     ┏━━━━━━━━━━━━━━━┻━━━━━━━━━━━━━━┓
   [0.60]: CAT                [0.40]: DOG
     ┃                              ┃
   [1.00]: SAT            ┏━━━━━━━━━┻━━━━━━━━┓
     ┃                   [0.50]: SAT       [0.50]: RAN
┏━━━━┻━━━━━━━┓             ┃                  ┃
HERE       ON            THERE              HERE
[0.33]    [0.67]         [1.00]             [1.00]

import Ngram from './ngram.js';

const ngram = new Ngram({
  input: 'The cat sat on the mat. The cat sat on the hat. The dog ran here.',
  size: 3  // trigram (default: 4)
});

// Suggest the most probable next word
ngram.suggestNextWord('cat sat');  // 'on'

// Complete a phrase up to N words or end-of-sentence
ngram.complete('The cat', 5);      // 'The cat sat on the mat'

API

Method	Returns	Description
new Ngram({ input, size? })	Ngram	Build from a text string. size defaults to 4 (4-gram).
complete(prefix, maxWords?)	string	Extends prefix with up to maxWords predicted words. Stops at sentence-ending punctuation or a prediction loop.
suggestNextWord(sequence, opts?)	string	Returns the single highest-probability next word. Pass { sanitize: true } to strip trailing punctuation.
setValue(text)	void	Replace the model with a new text corpus.

---

Levenshtein

Weighted Levenshtein distance between two strings, with separate costs for revisions, additions, and deletions. Used internally by the Trie for fuzzy search.

import levenshtein from './levenshtein.js';

levenshtein('court', 'cats', Infinity);  // weighted edit distance

The default weights are:

Operation	Cost
Revision (substitution)	1
Addition (insertion)	1
Deletion	2

A maximumDistance parameter enables early exit — if the length difference alone exceeds it, the function returns immediately without computing the full matrix.