strsimkit

Zero-dependency string similarity and distance algorithms — the maintained replacement for the archived string-similarity package.

Inspired by Python's textdistance (34 algorithms) and jellyfish, now in pure TypeScript with no runtime dependencies.

import { levenshteinSimilarity, jaroWinkler, bestMatch, soundex } from "strsimkit";

jaroWinkler("MARTHA", "MARHTA");           // 0.961
levenshteinSimilarity("kitten", "sitting"); // 0.571

bestMatch("appl", ["apple", "banana", "apricot"]);
// { bestMatch: "apple", rating: 0.889, index: 0 }

soundex("Robert") === soundex("Rupert"); // true — sound alike

Why strsimkit?

The dominant npm string-similarity package (string-similarity, 2.18M weekly downloads) was deprecated and archived in 2023 — yet it still has millions of weekly downloads because no maintained replacement existed. strsimkit fills that gap with a comprehensive, TypeScript-native, zero-dependency library.

Features

• Edit distance: Levenshtein (O(n) space), Damerau-Levenshtein (with transpositions)
• Jaro / Jaro-Winkler: industry-standard record linkage algorithms
• Token-based: Dice/bigram (Python difflib ratio), Jaccard, Overlap coefficient
• Hamming: bit-error distance, with padding for unequal lengths
• LCS: Longest Common Subsequence similarity + Longest Common Substring length
• Phonetic: Soundex, Double Metaphone, soundsLike() predicate
• bestMatch(): drop-in replacement for string-similarity's API
• rankMatches(): rate + sort all candidates, return top-N
• All similarity functions are normalized to [0, 1]
• Zero dependencies, TypeScript-first, ESM + CJS, tree-shakeable

Install

npm install strsimkit

Usage

Drop-in string-similarity replacement

import { bestMatch, levenshteinSimilarity } from "strsimkit";

// Was: stringSimilarity.findBestMatch(query, candidates)
const result = bestMatch("javascript", ["python", "javascript", "typescript"]);
result.bestMatch;  // "javascript"
result.rating;     // 1
result.index;      // 1

// Was: stringSimilarity.compareTwoStrings(a, b)
levenshteinSimilarity("hello", "helo"); // 0.889

Edit distance algorithms

import { levenshtein, levenshteinSimilarity, damerauLevenshtein } from "strsimkit";

levenshtein("kitten", "sitting");         // 3
levenshteinSimilarity("kitten", "sitting"); // 0.571

// Damerau-Levenshtein counts transpositions as 1 edit
damerauLevenshtein("abc", "acb"); // 1  (swap b↔c)
levenshtein("abc", "acb");        // 2  (substitute + substitute)

Jaro / Jaro-Winkler

import { jaro, jaroWinkler } from "strsimkit";

jaro("MARTHA", "MARHTA");        // 0.944
jaroWinkler("MARTHA", "MARHTA"); // 0.961  — higher for common prefix
jaroWinkler("DWAYNE", "DUANE");  // 0.840

Token-based similarity

import { dice, jaccard, overlap } from "strsimkit";

// Dice bigram — same as Python difflib SequenceMatcher.ratio()
dice("night", "nacht"); // 0.25

// Jaccard on character n-grams
jaccard("abc", "abd");          // 0.5   (bigrams)
jaccard("abc", "abd", 1);       // 0.667 (unigrams)

// Overlap coefficient
overlap("abc", "abcd");         // 1.0  (smaller set is subset)

Hamming distance

import { hamming, hammingPad, hammingSimilarity } from "strsimkit";

hamming("karolin", "kathrin"); // 3
hammingPad("abc", "abcde");    // 2  (pads shorter with null bytes)
hammingSimilarity("abc", "aXc"); // 0.667

LCS similarity

import { lcsLength, lcsSimilarity, longestCommonSubstring } from "strsimkit";

lcsLength("ABCBDAB", "BDCABA");    // 4
lcsSimilarity("hello", "hallo");   // 0.8
longestCommonSubstring("abcdef", "bcdfgh"); // 3 ("bcd")

Phonetic algorithms

import { soundex, doubleMetaphone, soundsLike } from "strsimkit";

soundex("Robert"); // "R163"
soundex("Rupert"); // "R163" — same code!
soundsLike("Robert", "Rupert"); // true

const [primary, secondary] = doubleMetaphone("Smith");

bestMatch and rankMatches

import { bestMatch, rankMatches, jaroWinkler } from "strsimkit";

// Best match with default (Levenshtein similarity)
bestMatch("hello", ["helo", "world", "hell"]);
// { bestMatch: "hell", rating: 0.8, index: 2 }

// Best match with custom algorithm
bestMatch("hello", ["helo", "jello"], jaroWinkler);

// Rank all candidates
rankMatches("night", ["nacht", "night", "knight"]);
// [{ candidate: "night", rating: 1 }, { candidate: "knight", rating: 0.857 }, ...]

API

| Export | Description | |--------|-------------| | levenshtein(a, b) | Edit distance (insert/delete/substitute). | | levenshteinSimilarity(a, b) | Normalized [0,1] Levenshtein. | | damerauLevenshtein(a, b) | Edit distance including transpositions. | | jaro(a, b) | Jaro similarity [0,1]. | | jaroWinkler(a, b, p?) | Jaro-Winkler similarity [0,1]. | | dice(a, b) | Sørensen-Dice bigram similarity [0,1]. | | jaccard(a, b, n?) | Jaccard n-gram similarity [0,1]. | | overlap(a, b, n?) | Overlap coefficient [0,1]. | | hamming(a, b) | Hamming distance (equal length required). | | hammingPad(a, b) | Hamming distance with padding. | | hammingSimilarity(a, b) | Normalized [0,1] Hamming. | | lcsLength(a, b) | Longest Common Subsequence length. | | lcsSimilarity(a, b) | LCS-based similarity [0,1]. | | longestCommonSubstring(a, b) | Length of longest common contiguous substring. | | soundex(s) | Soundex phonetic code (4 chars). | | doubleMetaphone(s) | [primary, secondary] Double Metaphone codes. | | soundsLike(a, b) | true if Soundex codes match. | | bestMatch(query, candidates, fn?) | Best candidate + score + index. | | rankMatches(query, candidates, fn?) | All candidates sorted by score desc. |

vs. alternatives

| Package | TypeScript | ESM | Algorithms | Phonetic | Maintained | |---------|-----------|-----|------------|----------|------------| | strsimkit | ✅ | ✅ | 12+ | ✅ | ✅ | | string-similarity | ❌ | ❌ | 1 | ❌ | ❌ archived 2023 | | natural | partial | ❌ | many | ✅ | ⚠️ heavy | | talisman | partial | ✅ | many | ✅ | ⚠️ last release 2022 |

Contributors ✨

Contributions of any kind are welcome! See the contributing guide and add yourself via @all-contributors please add @<username> for <contributions>.

License

MIT © trananhtung