NAME "Text::Ngramize" - Computes lists of n-grams from text. SYNOPSIS use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (normalizeText => 1); my $text = "This sentence has 7 words; doesn't it?"; dump $ngramizer-> (text => \$text); DESCRIPTION "Text::Ngramize" is used to compute the list of n-grams derived from the bytes, characters, or words of the text provided. Methods are included that provide positional information about the n-grams computed within the text. CONSTRUCTOR "new" use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (normalizeText => 1); my $text = ' To be.'; dump $ngramizer->getListOfNgrams (text => \$text); # dumps: # ["to ", "o b", " be", "be "] The constructor "new" has optional parameters that set how the n-grams are computed. "typeOfNgrams" is used to set the type of tokens used in the n-grams, "normalizeText" is used to set normalization of the text before the tokens are extracted, and "ngramWordSeparator" is the character used to join n-grams of words. "typeOfNgrams" typeOfNgrams => 'characters' "typeOfNgrams" sets the type of tokens to extract from the text to form the n-grams: 'asc' indicates the list of ASC characters comprising the bytes of the text are to be used, 'characters' indicates the list of characters are to be used, and 'words' indicates the words in the text are to be used. Note a word is defined as a substring that matches the Perl regular expression '\p{Alphabetic}+', see perlunicode for details. The default is 'characters'. "sizeOfNgrams" sizeOfNgrams => 3 "sizeOfNgrams" holds the size of the n-grams that are to be created from the tokens extracted. Note n-grams of size one are the tokens themselves. "sizeOfNgrams" should be a positive integer; the default is three. "normalizeText" normalizeText => 0 If "normalizeText" evalutes to true, the text is normalized before the tokens are extracted; normalization proceeds by converting the text to lower case, replacing all non-alphabetic characters with a space, compressing multiple spaces to a single space, and removing any leading space but not a trailing space. The default value of "normalizeText" is zero, or false. "ngramWordSeparator" ngramWordSeparator => ' ' "ngramWordSeparator" is the character used to separate token-"words" when forming n-grams from them. It is only used when "typeOfNgrams" is set to 'words'; the default is a space. Note, this is used to avoid having n-grams clash, for example, with bigrams the word pairs 'a aaa' and 'aa aa' would produce the same n-gram 'aaaa' without a space separating them. METHODS "getTypeOfNgrams" Returns the type of n-grams computed as a string, either 'asc', 'characters', or 'words'. use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (); dump $ngramizer->getTypeOfNgrams; # dumps: # "characters" "getSizeOfNgrams" Returns the size of n-grams computed. use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (); dump $ngramizer->getSizeOfNgrams; # dumps: # 3 "getListOfNgrams" The function "getListOfNgrams" returns an array reference to the list of n-grams computed from the text provided or the list of tokens provided by "listOfTokens". "text" text => ... "text" holds the text that the tokens are to be extracted from. It can be a single string, a reference to a string, a reference to an array of strings, or any combination of these. "listOfTokens" listOfTokens => ... Optionally, if "text" is not defined, then the list of tokens to use in forming the n-grams can be provided by "listOfTokens", which should point to an array reference of strings. An example using the method: use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (typeOfNgrams => 'words', normalizeText => 1); my $text = "This isn't a sentence."; dump $ngramizer->getListOfNgrams (text => \$text); # dumps: # ["this isn t", "isn t a", "t a sentence"] dump $ngramizer->getListOfNgrams (listOfTokens => [qw(aa bb cc dd)]); # dumps: # ["aa bb cc", "bb cc dd"] "getListOfNgramsWithPositions" The function "getListOfNgramsWithPositions" returns an array reference to the list of n-grams computed from the text provided or the list of tokens provided by "listOfTokens". Each item in the list returned is of the form "['n-gram', starting-index, n-gram-length]"; the starting index and n-gram length are relative to the unnormalized text. When "typeOfNgrams" is 'asc' the index and length refer to bytes, when "typeOfNgrams" is 'characters' or 'words' they refer to characters. "text" text => ... "text" holds the text that the tokens are to be extracted from. It can be a single string, a reference to a string, a reference to an array of strings, or any combination of these. "listOfTokens" listOfTokens => ... Optionally, if "text" is not defined, then the list of tokens to use in forming the n-grams can be provided by "listOfTokens", which should point to an array reference where each item in the array is of the form "[token, starting-position, length]", where "starting-position" and "length" are integers indicating the position of the token. An example using the method: use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (typeOfNgrams => 'words', normalizeText => 1); my $text = " This isn't a sentence."; dump $ngramizer->getListOfNgramsWithPositions (text => \$text); # dumps: # [ # ["this isn t", 1, 11], # ["isn t a", 7, 7], # ["t a sentence", 11, 13], # ] "getListOfNgramHashValues" The function "getListOfNgramHashValues" returns an array reference to the list of integer hash values computed from the n-grams of the text provided or the list of tokens provided by "listOfTokens". The advantage of using hashes over strings is that they take less memory and are theoretically faster to compute. With strings the time to compute the n-grams is proportional to their size, with hashes it is not since they are computed recursively. Also, the amount of memory used to store the n-gram strings grows proportional to their size, with hashes it does not. The disadvantage lies with hashing collisions, but these will be very rare. However, for small n-gram sizes hash values may take more time to compute since all code is written in Perl. "text" text => ... "text" holds the text that the tokens are to be extracted from. It can be a single string, a reference to a string, a reference to an array of strings, or any combination of these. "listOfTokens" listOfTokens => ... Optionally, if "text" is not defined, then the list of tokens to use in forming the n-grams can be provided by "listOfTokens", which should point to an array reference of strings. An example using the method: use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (typeOfNgrams => 'words', normalizeText => 1); my $text = "This isn't a sentence."; dump $ngramizer->getListOfNgramHashValues (text => \$text); # NOTE: hash values may vary across computers. # dumps: # [ # "4038955636454686726", # "5576083060948369410", # "6093054335710494749", # ] dump $ngramizer->getListOfNgramHashValues (listOfTokens => [qw(aa bb cc dd)]); # dumps: # ["7326140501871656967", "5557417594488258562"] "getListOfNgramHashValuesWithPositions" The function "getListOfNgramHashValuesWithPositions" returns an array reference to the list of integer hash values and n-gram positional information computed from the text provided or the list of tokens provided by "listOfTokens". Each item in the list returned is of the form "['n-gram-hash', starting-index, n-gram-length]"; the starting index and n-gram length are relative to the unnormalized text. When "typeOfNgrams" is 'asc' the index and length refer to bytes, when "typeOfNgrams" is 'characters' or 'words' they refer to characters. The advantage of using hashes over strings is that they take less memory and are theoretically faster to compute. With strings the time to compute the n-grams is proportional to their size, with hashes it is not since they are computed recursively. Also, the amount of memory used to store the n-gram strings grows proportional to their size, with hashes it does not. The disadvantage lies with hashing collisions, but these will be very rare. However, for small n-gram sizes hash values may take more time to compute since all code is written in Perl. "text" text => ... "text" holds the text that the tokens are to be extracted from. It can be a single string, a reference to a string, a reference to an array of strings, or any combination of these. "listOfTokens" listOfTokens => ... Optionally, if "text" is not defined, then the list of tokens to use in forming the n-grams can be provided by "listOfTokens", which should point to an array reference where each item in the array is of the form "[token, starting-position, length]", where "starting-position" and "length" are integers indicating the position of the token. An example using the method: use Text::Ngramize; use Data::Dump qw(dump); my $ngramizer = Text::Ngramize->new (typeOfNgrams => 'words', normalizeText => 1); my $text = " This isn't a sentence."; dump $ngramizer->getListOfNgramHashValuesWithPositions (text => \$text); # NOTE: hash values may vary across computers. # dumps: # [ # ["4038955636454686726", 1, 11], # ["5576083060948369410", 7, 7], # ["6093054335710494749", 11, 13], # ] INSTALLATION To install the module run the following commands: perl Makefile.PL make make test make install If you are on a windows box you should use 'nmake' rather than 'make'. AUTHOR Jeff Kubina COPYRIGHT Copyright (c) 2009 Jeff Kubina. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself. The full text of the license can be found in the LICENSE file included with this module. KEYWORDS information processing, ngram, ngrams, n-gram, n-grams, string, text SEE ALSO Encode, perlunicode, utf8