NAME

Lingua::TFIDF - Language-independent TF-IDF calculator.

VERSION

version 0.01

SYNOPSIS

  use Lingua::TFIDF;
  use Lingua::TFIDF::WordSegmenter::SplitBySpace;
  
  my $tf_idf_calc = Lingua::TFIDF->new(
    # Use a word segmenter for japanese text.
    word_segmenter => Lingua::TFIDF::WordSegmenter::SplitBySpace->new,
  );
  
  my $document1 = 'Humpty Dumpty sat on a wall...';
  my $document2 = 'Remember, remember, the fifth of November...';
  
  my $tf = $tf_idf_calc->tf(document => $document1);
  # TF of word "Dumpty" in $document1.
  say $tf->{'Dumpty'};  # 2, if you are referring same text as mine.
  
  my $idf = $tf_idf_calc->idf(documents => [$document1, $document2]);
  say $idf->{'Dumpty'};  # log(2/1) ≒ 0.693147
  
  my $tf_idfs = $tf_idf_calc->tf_idf(documents => [$document1, $document2]);
  # TF-IDF of word "Dumpty" in $document1.
  say $tf_idfs->[0]{'Dumpty'};  # 2 log(2/1) ≒ 1.386294
  # Ditto. But in $document2.
  say $tf_idfs->[1]{'Dumpty'};  # 0

DESCRIPTION

Quoting Wikipedia:

  tf–idf, short for term frequency–inverse document frequency, is a numerical statistic that is intended to reflect how important a word is to a document in a collection or corpus. It is often used as a weighting factor in information retrieval and text mining.

This module provides feature for calculating TF, IDF and TF-IDF.

MOTIVATION

There are several TF-IDF calculator modules in CPAN already, for example Text::TFIDF and Lingua::JA::TFIDF. So why I reinvent the wheel? The reason is language dependency: Text::TFIDF assumes that words in sentence are separated by spaces. This assumption is not true in most east asian languages. And Lingua::JA::TFIDF works only on japanese text.

Lingua::TFIDF solves this problem by separating word segmentation process from word frequency counting. You can process documents written in any languages, by providing appropriate word segmenter (see "CUSTOM WORD SEGMENTER" below.)

METHODS

new(word_segmenter => $segmenter)

Constructor. Takes 1 mandatory parameter word_segmenter.

CUSTOM WORD SEGMENTER

Although this distribution bundles some language-independent word segmenter, like Lingua::TFIDF::WordSegmenter::SplitBySpace, sometimes language-specifiec word segmenters are more appropriate. You can pass a custom word segmenter object to the calculator.

The word segmenter is a plain Perl object that implements segment method. The method takes 1 positional argument $document, which is a string or a reference to string. It is expected to return an word iterator as CodeRef.

Roughly speaking, given custom word segmenter will be used like:

  my $document = 'foo bar baz';
  
  # Can be called with a reference, like |->segment(\$document)|.
  # Detecting data type is callee's responsibility.
  my $iter = $word_segmenter->segment($document);
  
  while (defined(my $word = $iter->())) {
     ...
  }

idf(documents => \@documents)

Calculates IDFs. Result is returned as HashRef, which the keys and values are words and corresponding IDFs respectively.

tf(document => $document | \$document [, normalize => 0])

Calculates TFs. Result is returned as HashRef, which the keys and values are words and corresponding TFs respectively.

If optional parameter <normalize> is set true, the TFs are devided by the number of words in the $document. It is useful when comparing TFs with other documents.

tf_idf(documents => \@documents [, normalize => 0])

Calculates TF-IDFs. Result is returned as ArrayRef of HashRef. Each HashRef contains TF-IDF values for corresponding document.

SEE ALSO

Lingua::TFIDF::WordSegmenter::LetterNgram

Lingua::TFIDF::WordSegmenter::SplitBySpace

Lingua::TFIDF::WordSegmenter::JA::MeCab

AUTHOR

Koichi SATOH <sekia@cpan.org>

COPYRIGHT AND LICENSE

This software is Copyright (c) 2014 by Koichi SATOH.

This is free software, licensed under:

  The MIT (X11) License

cpanm Lingua::TFIDF

perl -MCPAN -e shell
install Lingua::TFIDF