CODEWITHSUNDEEP
searchtexthashpruning
Eric Pruitt
Eric Pruitt

Reputation: 1903

Numerical hash for comparing lexical similarity

Is there some form of hashing algorithm that produces similar numerical values for similar words? I imagine there would be a number of false positives, but it seems like something that could be useful for search pruning.

EDIT: Soundex is neat and may come in handy, but ideally, I want something that behave something like this: abs(f('horse') - f('hoarse')) < abs(f('horse') - f('goat'))

Upvotes: 3

Views: 699

Answers (4)

user703016
user703016

Reputation: 37945

What you are talking about is called Locality-sensitive Hashing. It can be applied to different types of input (images, music, text, positions in space, whatever you need).

Unfortunately (and despite searching) I couldn't find any practical implementation of an LSH algorithm for strings.

Upvotes: 2

Node
Node

Reputation: 3509

Checkout the Soundex algorithm on wikipedia, you haven't specified a language, but there are links to example implementations in multiple languages there. Obviously, this will give you a string hash thats the same for similar sounding words, and you want an integer, but you could then apply the string->integer hashing method they use in Boost.Hash.

Edit: To clarify, here is an example C++ implementation...

#include <boost/foreach.hpp>
#include <boost/functional/hash.hpp>

#include <algorithm>
#include <string>
#include <iostream>

char SoundexChar(char ch)
{
    switch (ch)
    {
        case 'B':
        case 'F':
        case 'P':
        case 'V':
            return '1';
        case 'C':
        case 'G':
        case 'J':
        case 'K':
        case 'Q':
        case 'S':
        case 'X':
        case 'Z':
            return '2';
        case 'D':
        case 'T':
            return '3';
        case 'M':
        case 'N':
            return '5';
        case 'R':
            return '6';
        default:
            return '.';
    }
}

std::size_t SoundexHash(const std::string& word)
{
    std::string soundex;
    soundex.reserve(word.length());

    BOOST_FOREACH(char ch, word)
    {
        if (std::isalpha(ch))
        {
            ch = std::toupper(ch);

            if (soundex.length() == 0)
            {
                soundex.append(1, ch);
            }
            else
            {
                ch = SoundexChar(ch);

                if (soundex.at(soundex.length() - 1) != ch)
                {
                    soundex.append(1, ch);
                }
            }
        }
    }

    soundex.erase(std::remove(soundex.begin(), soundex.end(), '.'), soundex.end());

    if (soundex.length() < 4)
    {
        soundex.append(4 - soundex.length(), '0');
    }
    else if (soundex.length() > 4)
    {
        soundex = soundex.substr(0, 4);
    }

    return boost::hash_value(soundex);
}

int main()
{
    std::cout << "Color = " << SoundexHash("Color") << std::endl;
    std::cout << "Colour = " << SoundexHash("Colour") << std::endl;

    std::cout << "Gray = " << SoundexHash("Gray") << std::endl;
    std::cout << "Grey = " << SoundexHash("Grey") << std::endl;

    return 0;
}

Upvotes: 0

Jonas Elfstr&#246;m
Jonas Elfstr&#246;m

Reputation: 31428

You could always try Soundex and see if it fits your needs.

Upvotes: 0

Andreas Jansson
Andreas Jansson

Reputation: 3297

The Soundex algorithm generates strings of keys corresponding to the phonemes in the input word. http://www.archives.gov/research/census/soundex.html

If you only want to compare similarity between strings, try Levenstein Distance. http://en.wikipedia.org/wiki/Levenshtein_distance

Upvotes: 3

Related Questions