Skip to content

Fast string matcher conceived for chat quick replies, voice interaction channels and multiple-choice scenario

License

Notifications You must be signed in to change notification settings

drksbr/quick-match

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

quick-match

Build Status NPM version js-standard-style stability-stable

Conversational interfaces are increasingly popular, Artificial intelligence, NLP/NLU solutions are at the forefront.
But, there is often the need for something simple, blazingly fast, and "offline" to solve the text matching problem.
This is a common issue with chat Quick Replies, multiple choice answers, and vocal interfaces (even less reliable).
Quick Match provides a toolkit to address all these problems in the best possible way.

Do you have a good idea or want to make the matching algorithm even more efficient? Collaborate!

Usage

Install

npm i -s quick-match

Simple initialization with default algorithm:

const { QuickMatch } = require('quick-match')

const qm = new QuickMatch()

const userInput = 'I want a pizza'
const candidates = ['Free hot dog here', 'Pizza for sale', 'Rent your cola']

const { bestCandidateIdx } = qm.run(userInput, candidates) // 1

Available algorithms

  • Dice's coefficient (max score => best result)
const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({ algorithm: 'dice' })

const input = 'I want a pizza'
const qr = ['Free hot dog here', 'Pizza for sale', 'Rent your cola']

const { candidates, bestCandidateIdx, maxScore } = qm.run(input, qr)

// candidates
// [
//   {
//     text: 'Free hot dog here',
//     keywords: [],
//     score: 0.06896551724137931,
//     stemmed: [ 'free', 'here' ],
//     intersections: []
//   },
//   {
//     text: 'Pizza for sale',
//     keywords: [],
//     score: 0.3076923076923077,
//     stemmed: [ 'pizza', 'sale' ],
//     intersections: [ 'pizza' ]
//   },
//   {
//     text: 'Rent your cola',
//     keywords: [],
//     score: 0.15384615384615385,
//     stemmed: [ 'rent', 'your', 'cola' ],
//     intersections: []
//   }
// ]

// bestCandidateIdx: 1
// maxScore: 0.3076923076923077
  • Levenshtein distance (min score => best result)
const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({ algorithm: 'levenshtein' })

const input = 'I want a pizza'
const qr = ['Free hot dog here', 'Pizza for sale', 'Rent your cola']

const { candidates, bestCandidateIdx, minScore } = qm.run(input, qr)

// candidates
// [
//   {
//     text: 'Free hot dog here',
//     keywords: [],
//     score: 15,
//     stemmed: [ 'free', 'here' ],
//     intersections: []
//   },
//   {
//     text: 'Pizza for sale',
//     keywords: [],
//     score: 12,
//     stemmed: [ 'pizza', 'sale' ],
//     intersections: [ 'pizza' ]
//   },
//   {
//     text: 'Rent your cola',
//     keywords: [],
//     score: 12,
//     stemmed: [ 'rent', 'your', 'cola' ],
//     intersections: []
//   }
// ]

// bestCandidateIdx: 1
// minScore: 12

Stemming

You can also enable them stemming, applying the algorithms only on the stem (root) of the words.
This, sometimes, is useful to reduce the noise.

const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({
  algorithm: 'dice',
  enableStemming: true,
  stemming: {
    language: 'English',
    // (optional) Shorter words than this number BEFORE stemming are removed
    minPreStemmingLength: 3,
    // (optional) Shorter words than this number AFTER stemming are removed
    minPostStemmingLength: 4
  }
})

const input = 'I discussed about food'
const qr = [
  { text: 'Discussing food' },
  { text: 'Eating and running' }
]

const { stemmedText, bestCandidateIdx } = qm.run('i have discussed about food', qr)

// stemmedText: [ 'have', 'discuss', 'about', 'food', 'drink' ]
// bestCandidateIdx: 0

Available languages:

  • Arabic
  • Danish
  • Dutch
  • English
  • French
  • German
  • Greek
  • Hungarian
  • Italian
  • Portuguese
  • Romanian
  • Russian
  • Spanish
  • Swedish
  • Tamil
  • Turkish

Words intersections

Another useful feature is the possibility to have the intersections among user input and possible candidates. This can be used further to better assess the fittest candidate.

const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({
  algorithm: 'dice',
  enableStemming: true,
  stemming: { language: 'English', minPreStemmingLength: 4, minPostStemmingLength: 4 }
})

const input = 'I have discussed about mealing and foot'
// As you can see, the candidates can be only strings or object in the format { text: string, keywords: [string] } to improve the matching with relatex keyworkds
const qr = [
  { text: 'Discussing food', keywords: ['eating', 'meal'] },
  { text: 'Eating and running', keywords: ['jogging', 'footing'] }
]

const { candidates, maxIntersections, maxIntersectionsCandidateIdx, bestCandidateIdx } = qm.run(input, qr)
// candidates
// [
//   {
//     text: 'Discussing food',
//     keywords: [ 'eating', 'meal' ],
//     score: 0.4230769230769231,
//     stemmed: [ 'discuss', 'food', 'meal' ],
//     intersections: [ 'discuss', 'meal' ]
//   },
//   {
//     text: 'Eating and running',
//     keywords: [ 'jogging', 'footing' ],
//     score: 0.2545454545454545,
//     stemmed: [ 'foot' ],
//     intersections: [ 'foot' ]
//   }
// ]

// maxIntersections: 2
// maxIntersectionsCandidateIdx: 0
// bestCandidateIdx: 0

Usage of numbers

It sometimes useful, especially in chat and voice interaction to recognize digits as user's answer

Simple digit matching

const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({
  numbers: {
    enableDigits: true,
    maxDigit: 5
  }
})
const qr = ['foo', 'bar', 'zoo']
const { numberMatch, numberMatchType, bestCandidateIdx } = qm.run('1', qr)

// numberMatch: true
// numberMatchType: 'digit'
// bestCandidateIdx: 0

Simple cardinal matching

const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({
  numbers: {
    enableCardinals: true,
    cardinals: ['uno', 'due', 'tre'] // Specify your custom cardinals based on your language
  }
})
const qr = ['foo', 'bar', 'zoo']
const res = qm.run('la due', qr) // It tries to get the number even if other small "noise" words

// numberMatch: true
// numberMatchType: 'cardinal'
// bestCandidateIdx: 1

Simple ordinal matching

const { QuickMatch } = require('quick-match')
const qm = new QuickMatch({
  numbers: {
    enableOrdinals: true,
    ordinals: ['prima', 'seconda', 'terza'] // Specify your custom ordinals based on your language
  }
})
const qr = ['foo', 'bar', 'zoo']
const res = qm.run('seconda scelta', qr)

// numberMatch: true
// numberMatchType: 'ordinal'
// bestCandidateIdx: 1

Lots of possibilities

A lot of customization with options for every detail:

const { QuickMatch } = require('quick-match')

const qm = new QuickMatch({
  algorithm: 'dice',
  enableStemming: true,
  stemming: {
    language: 'English',
    minPreStemmingLength: 4,
    minPostStemmingLength: 4
  },
  enableAlgorithmOnKeywords: false,
  numbers: {
    enableDigits: true,
    enableCardinals: true,
    enableOrdinals: true,
    maxDigit: 10,
    maxWordsEnablingNumbers: 2,
    cardinals: [
      'one', 'two', 'three', 'four', 'five',
      'six', 'seven', 'eigth', 'nine', 'ten'
    ],
    ordinals: [
      'first', 'second', 'third', 'fourth', 'fifth',
      'sixth', 'seventh', 'eighth', 'ninth', 'tenth'
    ]
  },
  limits: {
    minLengthCandidate: 3,
    maxCandidateWords: 5
  },
  weightIntersectionMultiplier: 1
})

// Declare your input text, candidates and run the algorithm
const userInput = 'I want a pizza'
const candidates = [
  { text: 'Free hot dog here', keywords: ['hot dog', 'free'] },
  { text: 'Pizza for sale', keywords: ['pizza', 'margherita'] },
  { text: 'Rent your cola', keywords: ['coke', 'cola'] }
]

const { bestCandidateIdx } = qm.run(userInput, candidates) // 1

Result output format:

{
  "algorithm": "dice",
  "minScore": 0.06451612903225806,
  "maxScore": 0.2857142857142857,
  "maxIntersections": 1,
  "candidates": [
    {
      "text": "I want hot-dog here",
      "keywords": [
        "hot-dog",
        "free"
      ],
      "score": 0.24242424242424243,
      "stemmed": [
        "want",
        "here",
        "hot-dog",
        "free"
      ],
      "intersections": [
        "want"
      ]
    },
    {
      "text": "Pizza for sale",
      "keywords": [
        "pizza",
        "margherita"
      ],
      "score": 0.2857142857142857,
      "stemmed": [
        "pizza",
        "sale",
        "pizza",
        "margherita"
      ],
      "intersections": [
        "pizza"
      ]
    },
    {
      "text": "Renting your cola",
      "keywords": [
        "coke",
        "cola"
      ],
      "score": 0.06451612903225806,
      "stemmed": [
        "rent",
        "your",
        "cola",
        "coke",
        "cola"
      ],
      "intersections": []
    }
  ],
  "text": "I wanted a pizza",
  "stemmedText": [
    "want",
    "pizza"
  ],
  "minCandidateIdx": 2,
  "maxCandidateIdx": 1,
  "maxIntersectionsCandidateIdx": 0,
  "bestCandidateIdx": 1,
  "bestCandidate": {
    "text": "Pizza for sale",
    "keywords": [
      "pizza",
      "margherita"
    ],
    "score": 0.2857142857142857,
    "stemmed": [
      "pizza",
      "sale",
      "pizza",
      "margherita"
    ],
    "intersections": [
      "pizza"
    ]
  }
}

About

Fast string matcher conceived for chat quick replies, voice interaction channels and multiple-choice scenario

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • JavaScript 100.0%