Calculate String Similarity

Overview

You send two strings and name exactly one metric, and we return a score from 0 to 1 rounded to four decimals, the raw edit distance where the metric has one, the set arithmetic where it works on sets, and the lengths after normalisation so a surprising score is explainable. Seven metrics are offered and each is defined in the documentation: LEVENSHTEIN and DAMERAU_LEVENSHTEIN over code points, JARO and JARO_WINKLER with Winkler's published 0.7 prefix threshold, and JACCARD_TOKEN, JACCARD_SHINGLE and DICE_SHINGLE over token and word-shingle sets. There is no default metric, because a similarity score that does not say which metric produced it is not an answer. The cost budget is chosen by the metric, so a linear metric is never refused at a quadratic price, and a pair too expensive to compare is refused before the first cell rather than sampled.

One platform should have one Levenshtein and one Jaro-Winkler, so two skills scoring the same pair never disagree about the number.

When to use it

Scope

Non-goals

• Does not cluster records — data.detect_duplicates does that
• Does not choose a metric for you
• Does not decide whether a score means the same thing
• Does not compare more than one pair per call

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"data.calculate_similarity","input":{"a":"Jonathon","b":"Jonathan","method":"LEVENSHTEIN"}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("data.calculate_similarity", {"method":"LEVENSHTEIN","a":"Jonathon","b":"Jonathan"});

Errors

Other limits

Limitations

• Does not cluster records — data.detect_duplicates does that
• Does not choose a metric for you
• Does not decide whether a score means the same thing
• Does not compare more than one pair per call
• Deduplicating a whole table pair by pair
• Treating a score as a probability that two records are the same

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "data.calculate_similarity",
  "input": {
    "method": "LEVENSHTEIN",
    "a": "Jonathon",
    "b": "Jonathan"
  }
}

Examples

{
  "method": "LEVENSHTEIN",
  "a": "Jonathon",
  "b": "Jonathan"
}

{
  "score": 0.875,
  "distance": 1,
  "method": "LEVENSHTEIN",
  "normalization": {
    "stripPunctuation": false,
    "caseFold": false,
    "unicodeForm": "NONE",
    "collapseWhitespace": false
  },
  "work": {
    "projected": 64,
    "budget": 25000000
  },
  "lengths": {
    "b": 8,
    "a": 8
  }
}

{
  "method": "JACCARD_TOKEN",
  "a": "red cotton shirt",
  "b": "cotton shirt red"
}

{
  "score": 1,
  "sets": {
    "union": 3,
    "intersection": 3,
    "aSize": 3,
    "bSize": 3
  },
  "method": "JACCARD_TOKEN",
  "normalization": {
    "tokenPattern": "UNICODE_WORD",
    "unicodeForm": "NONE",
    "collapseWhitespace": false,
    "caseFold": false,
    "stripPunctuation": false
  },
  "work": {
    "projected": 6,
    "budget": 25000000
  },
  "lengths": {
    "b": 16,
    "a": 16
  }
}

Recommended use cases

• Comparing two candidate spellings of one name
• Scoring a proposed match before a human confirms it
• Measuring how far a rewritten text moved from the original

Avoid for

• Deduplicating a whole table pair by pair
• Treating a score as a probability that two records are the same

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags