fast-fuzzy

Fast fuzzy-search utility

methodology

fast-fuzzy is a tiny, lightning-quick fuzzy-searching utility. The ranking algorithm is a modification of levenshtein distance proposed by Peter H. Sellers (paper). fast-fuzzy also uses the damerau-levenshtein distance by default, which, compared to normal levenshtein, punishes transpositions less.

Inputs are normalized before search. Normalization consists of standard utf8-normalization, optionally taking the lowercase of a string, optionally removing non-word characters, and optionally flattening/trimming whitespace. Graphemes, such as conjoined emoji 👨‍👩‍👧‍, are treated as single characters.

Inputs are scored from 0 to 1, where a higher score indicates a closer match. When searching, results are returned in descending order of score. Ties are broken by favoring the candidate whose length is closest to the length of the search term. This causes matches which are closer to exact full string matches to be effectively ranked higher in the case of a tie. Ties in length difference are broken by insertion order.

Lists of candidates are stored in a trie internally, which avoids doing redundant work on candidates with common prefixes. Additionally, when a subtree of the trie can be determined to have no string which could possibly score >= the threshold, the entire subtree is skipped. This significantly improves search times compared to a bruteforce search.

While the default options are to use Damerau and Sellers (transposition-friendly substring search), either of these options can be opted out of if the need arises.

exports

name	description	signature
fuzzy	fuzzy ranking algorithm; returns match strength	(term, candidate, options?) => score
search	for one-off searches; returns a sorted array of matches	(term, candidates, options?) => matches
Searcher	for searching the same set of candidates multiple times; caches the constructed trie1	N/A

¹ it is recommended that you use a Searcher when searching the same set multiple times. search will create a new trie every time, and while this is relatively cheap, it can have an impact on responsiveness if you intend to update search results in real time, i.e. while typing.

Searcher methods

name	description	signature
constructor	supply the options and initial list of candidates	(candidates?, options?) => searcher
add	add new candidates to the list	(...candidates) => void
search	perform a search against the instance's candidates	(term, options?) => matches2

² allows overriding the threshold, returnMatchData, and useDamerau options

options

Searcher and search both take an options object for configuring behavior.

option	type	description	default
keySelector	Function	selects the string(s)3 to search when candidates are objects	s => s
threshold	Number	the minimum score that can be returned	.6
ignoreCase	Bool	normalize case by calling toLower on input and pattern	true
ignoreSymbols	Bool	strip non-word symbols4 from input	true
normalizeWhitespace	Bool	normalize and trim whitespace	true
returnMatchData	Bool	return match data5	false
useDamerau	Bool	use damerau-levenshtein distance	true
useSellers	Bool	use the Sellers method for substring matching	true

³ if the keySelector returns an array, the candidate will take the score of the highest scoring key.

⁴ `~!@#$%^&*()-=_+{}[]\|\;':",./<>?

⁵ in the form {item, original, key, score, match: {index, length}}. Match index and length are in terms of the original, non-normalized string. Also note that match will be undefined if useSellers is false.

fuzzy accepts a subset of these options (excluding keySelector and threshold) with the same defaults.

examples

You can call fuzzy directly to get a match score for a single string

const {fuzzy} = require("fast-fuzzy");

fuzzy("hello", "hello world"); //returns 1
fuzzy("word", "hello world"); //returns .75

//pass in custom options
fuzzy("hello world", "hello  world"); //returns 1
fuzzy("hello world", "hello  world", {normalizeWhitespace: false}); //returns .90909090...

Use search to search a list of strings or objects

const {search} = require("fast-fuzzy");

search("abc", ["def", "bcd", "cde", "abc"]); //returns ["abc", "bcd"]

//pass in a keySelector to search for objects
search(
    "abc",
    [{name: "def"}, {name: "bcd"}, {name: "cde"}, {name: "abc"}],
    {keySelector: (obj) => obj.name},
);
//returns [{name: "abc"}, {name: "bcd"}]

//pass returnMatchData to receive the matchData for each result
search("abc", ["def", "bcd", "cde", "abc"], {returnMatchData: true});
/* returns [{
    item: 'abc', original: 'abc', key: 'abc', score: 1,
    match: {index: 0, length: 3},
}, { 
    item: 'bcd', original: 'bcd', key: 'bcd', score: 0.6666666666666667,
    match: {index: 0, length: 2},
}] */

Use Searcher in much the same way as search

const {Searcher} = require("fast-fuzzy");

const searcher = new Searcher(["def", "bcd", "cde", "abc"]);
searcher.search("abc"); //returns ["abc", "bcd"]

//options are passed in on construction
const anotherSearcher = new Searcher(
    [{name: "thing1"}, {name: "thing2"}],
    {keySelector: (obj) => obj.name},
);

//some options can be overridden per call
searcher.search("abc", {returnMatchData: true});
/* returns [{
    item: 'abc', original: 'abc', key: 'abc', score: 1,
    match: {index: 0, length: 3},
}, { 
    item: 'bcd', original: 'bcd', key: 'bcd', score: 0.6666666666666667,
    match: {index: 0, length: 2},
}] */