Skip to content

Latest commit

 

History

History
132 lines (95 loc) · 6.01 KB

README.md

File metadata and controls

132 lines (95 loc) · 6.01 KB

Detection of near-duplicate documents

When you have lots of digital documents you would like to know how many duplicates or near-duplicates exist on your storage devices, because you want to clean up space, detect plagiarism or avoid include documents with different versions. Duplicate documents: When you need to find documents that are exact copies, these are easy to find - just calculate a checksum and compare with others. Near-duplicate: documents differ from each other in a very small portion, could be useful for a document life cycle or if you want to find documents which basically are different versions of the same document.

This software is near-duplicate detection system and uses Charikar's fingerprinting technique and Rabin-Hash function.

#Charikars algorithm implementation

##How can detect near-duplicates?

To detect near-duplicates this software uses the Charikar's fingerprinting technique, this means characterizing each document with a unique 64-bit vector, like a fingerprint. To determine whether two documents are Near-duplicates, we have to compare their fingerprints.

To do this we use two algorithms, the algorithm developed by Moses Charikar and the Hamming distance algorithm, which allows us to measure the similarity between two vectors of n bits.

##What is Charikar's algorithm?

  • Characterization of the document
  • Apply hash functions to the characteristics
  • Obtain fingerprint
  • Apply vector comparison function: Are (Doc1, doc2) near-duplicate? Hamming-distance (fingerprint (doc1), fingerprint (doc2)) = k

##What means characterize documents?

Characterizing Document is what things makes a document different from the others, for example number of words, number of paragraphs, punctuation, sentences, etc ...

It is up to us to define what is the most appropriate feature to generate the fingerprint.

One way to characterize is using a technique known as shingling. This technique consist to split documents in Shingles, set of words grouped in n and n, also called n-gram grouping.

For example, shingling in 3-grams: Phrase: detecting near duplicates of documents 3-grams: (detection of documents, documents near, near duplicates documents).

To further characterize the document, we remove stop-words (words without meaning) and do stemming on every word (extract the root of the word), previously we will detect the language.

So the above example would be: Phrase: detecting near duplicates of documents Stop-words: detecting near duplicates documents Stemming: near duplicate document detection 3-grams: Document detect near, near duplicate document

A singly extracted features apply a hash function to each of them.

We will apply the Rabin-Hash function, exists other functions.

There are many hash function that can be used. We decided to use the hash-Rabin because it very quickly.

For the phrase "Detecting near duplicates" we get the following 64-bit hash.

> Detecting - 0000000000000000000000000000000000000000000110100111110100001011
> Near - 0000000000000000000000000000000000000000000011110000000000010111
> Duplicates - 0000000000000000000000000000000001111101100000110110100101111111

##How do you calculate a fingerprint?

We start from a n-bit (despends on hash algorithm) vector with all 0, for each hash obtained from the documen we have to apply the following function: If the n-th bit vector is 1, increases resulting vector 1, but if its (<= 1) decreases in 1. example:

Init Vector  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
Detecting    0   1   1   0   1   0   0   1   1   1   1   1   0   1   0   0   0   0   1   0   1   1

Vector res. -1   1  1  -1  1  -1  -1   1   1   1   1   1  -1   1  -1  -1  -1  -1  1 -1   1   1
Near         0   0  1   1   1  1   0   0   0   0   0   0   0   0   0   0   0   1  0  1   1   1

Vector res.  -2  1   2   0   2   0  -2  0   0   0   0   0   -2  0   -2  -2  -2   0 -1  0   2   2  
Duplicates    1  1   0   0   0   0   0  1   1   0   1   1    0  1    0   0   1   0  1  1   1   1

Vector res.  -1  2   1  -1   1  -1  -3  1   1  -1   1   1   -3   1  -3  -3  -1  -1  0  1   3   3   

Once you have applied all the characterizations in hash format, generate the fingerprint as follows: If the resulting vector is negative or 0, the fingerprint in that position is 0, otherwise it would be 1. In this example the fingerprint be as follows:

> Result Vector3.  -1  2  1  -1  1  -1  -3   1  1  -1   1  1   -3   1   -3   -3  -1  -1  0 1 3 3   
> Fingerprint       0  1  1   0  1   0   0   1  1   0   1  1    0   1    0    0   0   0  0 1 1 1

#Hamming distance.

Algorithm to calculate the similarity between two documents, there are more, but we'll use this. Basically what this algorithm does is: given two vectors of n bits, several changes must be made equal to one another, this number will be called K-Hamming distance.

#When two document are Near-duplicates ?

We consider that two documents are near-duplicates when the K-Hamming distance resulting from applying the fingerprint is <= 3.

#Usage

After compiling the source code, you will have a file called dedup.jar. This utility needs several parameters: [file1] File to compare 1 [file2] File to compare 2 [max grams] n-grams number max. [min gram] n-grams number min

java -jar dedup.jar [file1] [file2]  max [max ngrams] min [min ngrams]
java -jar dedup.jar text_sample_2.txt text_sample_1.txt  max 3 min 3

#Output

This is an output sample.

Near duplicates detector
Charika simhash, with hashRabin and Hamming distance.
Token ngrams 3-3.
time taken to load all available language-models: 0.086s
time taken to create language-model from input: 0.018s
time taken to effectively determine the language: 0.011s
Loading English Steemer.
201 millisegons
time taken to load all available language-models: 0.048s
time taken to create language-model from input: 0.0040s
time taken to effectively determine the language: 0.0030s
Hamming-distance:0
Jaccard-Coeficient:100.0%
IS NEAR-DUPLICATE, COULD BE THE SAME DOCUMENTS.
287 millisegons