node-simhash-mod

64 bits version now

warning, much slower

A simple command line tool for comparing text files using the simhash algorithm and contrasting with the jaccard index.

Almost pure fork of node-simhash, by Scott Horn:

• Patches log4js issue by setting a forced version of log4js
• Cleans French diacritics
• getDistanceReport helper function

References

• Near duplicate detection (moz.com)
• Near duplicate detection (Jonathan Koren)

Installation

If you have just clone this like then run the following

npm install
npm link

Command line tool usage

Using node

simhash file1.txt file2.txt

simhash https://file.com/page1.html https://file.com/page2.html

Using lib

var simhash = require('node-simhash-mod');
 
simhash.compare(string1, string2);
 

Methods

.summary(file1, file2)

Compare two text strings using both simhash and jaccard index and print a summary

.compare(file1, file2)

Compare two text strings using both simhash and jaccard index

.hammingWeight(number)

Count the binary ones in a number.

.shingles(string, words_per_single=2)

Convert string to set of shingles using the default of 2 words per shingle and tokenize using the natural libraries default tokenizer.

.jaccardIndex(string1, string2)

Compare two strings by tokeniseing and then compare the intersection of shingles to the union of shingles.

.createBinaryString(number)

Print a 32-bit number as a binary string of 32 characters

.shingleHashList(set)

Convert a set of shingles to a set of crc-32 hashes.

Distance report

Often you have a list of strings, and what to check how close they are each from other.

getDistanceReport will produce a JSON report containing, for each text, the closest ones.

Parameters are the following:

• an array of textual objects; each object must have a text property containing its string, and a simhash property with the hash already calculated; feel free to put other properties typically an ID
• the maximal acceptable similarity: if the similarity between two strings is greater than this threshold, then it will be added in the list of the closest ones; use 0.8 for instance to only trigger when texts are 80% different or less
• the maximum number of closest strings to be given in the output (only the most close ones will be given)

The output is an array of objects:

• for: reference to the textual object
• closestOnes: an array with the closes elements; each object points to an element (with property) and gives the distance (difference property)