kocos.pl - Find the Kth order co-occurrences of a word
This program finds the Kth order co-occurrences of a given word.
Co-occurrences are the words which occur together in the same context. All words which co-occur with a given target word are called its co-occurrences. The concept of 2nd order co-occurrences is explained in the paper Automatic word Sense Discrimination [Schutze98]. According to this paper, the words which co-occur with the co-occurring words of a target word are called as the 2nd order co-occurrences of that word.
So with each increasing order of co-occurrences, we introduce an extra level of indirection and find words co-occurring with the previous order co-occurrences.
We generalize the concept of 2nd order co-occurrences from [Schutze98] to find the Kth order co-occurrences of a word. These are the words that co-occur with the (K-1)th order co-occurrences of a given target word.
We have also found [Niwa&Nitta94] to be related to kocos. While we do not exactly reimplement the co-occurrence vectors they propose, we feel that kocos is at least similar in spirit.
Usage: kocos.pl [OPTIONS] BIGRAM
Specify the BIGRAM file name on the command line after the program name and options (if any) as shown in the usage note.
BIGRAM should be a bigram output(normal or extended) created by NSP programs - count.pl, statistic.pl or combig.pl. When count.pl and statistic.pl are run for creating bigrams (--ngram set to 2 or not specified), the programs list the bigrams of all words which co-occur together(in certain window). So we can say that if a bigram 'word1<>word2<>' is listed in the output of count.pl or statistic.pl program, it means that the words word1 and word2 are the co-occurrences of each other.
In general you may want to consider the use of stop lists (--stop option in count.pl) to remove very common words such as "the" and "for", and also eliminate low frequency bigrams (--remove option in count.pl). The stop list is particularly important as high frequency words such as "the" or "for" will co-occur with many different words, and greatly expand the search needed to find kth order co-occurrences.
If you want to run kocos.pl on a source file not created by either count or statistic program of this package, just make sure that each line of BIGRAM file will list two words WORD1 and WORD2 as WORD1<>WORD2<> The program minimally requires that there are exactly two words and they are separated by delimiter '<>' with an extra delimiter '<>' after the second word. So you may convert any non NSP input to this format where two words occurring in the same context are '<>' separated and provide it to kocos.
Controlling scope of the context
You may like to call two words as co-occurrences of each other if they occur within a specific distance from each other. We encourage in this case that you use --window w option of NSP program count.pl while creating a BIGRAM file. This will create bigrams of all words which co-occur within a distance w from each other. Thus --window w sets the maximum distance allowed between two words to call them co-occurrences of each other.
Note that if the --window option is not used while creating BIGRAM input for kocos, only those words which come immediately next to each other will be considered as the co-occurrences (default window size being 2 for bigrams).
With this option, the target WORD whose kth order co-occurrences are to be found can be directly specified on the command line.
e.g. kocos.pl --literal line test.input will find the 1st order co-occurrences (by default) of the word 'line' using Bigrams listed in file test.input.
kocos.pl --literal , --order 3 test.input will find 3rd order co-occurrences of ',' from file test.input.
With this option, target word can be specified using Perl regular expression/s. The regex/s should be written in a file and multiple regex/s should either appear on separate lines or should be Perl 'OR' (|) separated.
We provide this option to allow user to specify various morphological variants of the target word e.g. line, lines, Line,Lines etc.
e.g. (1) let test.regex contains a regular expression for target word which is - /^[Ll]ines?$/
To use this for finding kocos, run kocos.pl with command
kocos.pl --regex test.regex --order K test.input
(2) To find say 2nd order co-occurrences of any general target word which occurs in Data in <head> tags like Senseval Format, we use a regular expression /^<head.*>\w+</head>$/ in our regex file say test.regex and run kocos.pl using command
kocos.pl --regex test.regex --order 2 eng-lex-sample.training.xml
(3) To find 3rd order co-occurrences of any word that contains period '.' run kocos.pl using
kocos.pl --literal . --order 3 test.input
Or write a regex /\./ in file say test.regex and run kocos using
kocos.pl --regex test.regex --order 3 test.input
(4) To find 2nd order co-occurrences of all words that are numbers, write a regex like /^\d+$/ to a regexfile say test.regex and run kocos using,
kocos.pl --regex test.regex --order 2 test.input
Note: writing a regex /\d+/ will also match words like line20.1.cord, or art%10.fine456 that include numbers.
Regex/s that should exactly match as target words should be delimited by ^ and $ as in /^[Ll]ines?$/. Specifying something like /[Ll]ines?/ will match with 'incline'.
Note - The program kocos.pl requires that the target word is specified using either of the options --literal or --regex
If the value of K is specified using the command line option --order K, kocos.pl will find the Kth order co-occurrences of the target word. K can take any integer value greater than 0. If the value of K is not specified, the program will set K to 1 and will simply find the co-occurrences of the target (the word co-occurrence generally means first order co-occurrences).
To see a detailed report of how each Kth order co-occurrence is reached as a sequence of K words, specify the name of a TRACEFILE on the command line using --trace TRACEFILE option.
TRACEFILE will show the chains of K+1 words where the first word is the TARGET word and every ith word in the chain is a (i-1)th order co-occurrence of target which co-occurs with (i-1)th word in the chain. So a chain of K+1 words,
TARGET->COC1->COC2->COC3....->COCK-1->COCK
shows that COC1 is a first order co-occurrence of the TARGET.
COC2 is a second order co-occurrence such that COC2 co-occurs with COC1 which in turn co-occurs with the TARGET. COC3 is a third order co-occurrence such that COC3 co-occurs with COC2 which in turn co-occurs with COC1 which co-occurs with TARGET.
and so on......
This option will display the help message.
This option will display version information of the program.
The program will display a list of Kth order co-occurrences to standard output such that each co-occurrence occurs on a separate line and is followed by '<>' (just to be compatible with other programs in NSP).
Note that the output of kocos.pl could be directly used by the program nsp2regex of the SenseTools Package (by Satanjeev Banerjee and Ted Pedersen) to convert Senseval data instances into feature vectors in ARFF format where our Kth order co-occurrences are used as features.
For more information on SenseTools you can refer to its README: http://www.d.umn.edu/~tpederse/sensetools.html
IMPORTANT NOTE
If there are some kth order co-occurrences which are also the ith order co-occurrences (0<i<k) of the target word, program kocos.pl will not display them as the Kth order co-occurrences. kocos.pl displays only those words as Kth order co-occurrences whose minimum distance from target word is K in the co-occurrence graph. [Co-occurrence graph shows a network of words where a word is connected to all words it co-occurs with.]
(a) Using default value of order To find the (1st order) co-occurrences of a word 'line' from the BIGRAM file test.input, run kocos.pl using the following command. kocos.pl --literal line test.input
(b) Using option order To find the 2nd order co-occurrences of a word 'line' from the BIGRAM file test.input, run kocos.pl using the following command. kocos.pl --literal line --order 2 test.input
(c) Using the trace option To see how the 4th order co-occurrences of a word 'line' is reached as a sequence of words which form a co-occurrence chain, run kocos.pl using the following command. kocos.pl --literal line --order 4 --trace test.trace test.input
(d) Using a Regex to specify the target word To find Kth order co-occurrences of a target word 'line' which is specified as a Perl regular expression say /^[Ll]ines?$/ in a file test.regex, run kocos.pl using kocos.pl --regex test.regex --order K test.input
(e) Using a generic Regex for Data like Senseval-2, To find 2nd order co-occurrences of a target word that occurs in <head> tags in the data file eng-lex-sample.training.xml, use a regular expression like /<head>\w+</head>/ from a file say test.regex, and run kocos.pl using kocos.pl --regex test.regex --order 2 test.input
(a) Create a BIGRAM file using programs count.pl, statistic.pl or combig.pl of NSP Package. (b) Use --window W option of program count.pl to specify the scope of the context. Any word that occurs within a distance W from a target word will be treated as its co-occurrence. (c) Use either --literal or --regex option to specify the target word. We recommend use of regex support to detect forms of target word other than its base form.
In all the following examples, we assume that the input comes from the file test.input and word 'line' is a target word.
test.input => ---------------- print<>in<> | print<>line<> | text<>the<> | text<>line<> | file<>the<> | file<>in<> | line<>file | ----------------
(Note that test.input doesn't look like a valid count/statistic output because kocos.pl will minimally require two words WORD1 and WORD2 separated by '<>' with an extra '<>' after WORD2 as described in Section 3.1 of this README)
(a) The 1st order co-occurrences of word 'line' can be found by running kocos.pl with either of the following commands -
kocos.pl --literal line test.input OR kocos.pl --order 1 --literal line test.input
This will display the co-occurrences of 'line' to standard output as shown below in the box.
-------- text<> | file<> | print<>| --------
This is because the program finds the bigrams
print<>line<> text<>line<> line<>file<>
where word 'line' co-occurs with the words print, text and file which become the 1st order co-occurrences.
(b) The 2nd order co-occurrences of word 'line' can be found by running kocos.pl with the following command - kocos.pl --literal line --order 2 test.input
This will display the 2nd order co-occurrences of 'line' to standard output as shown below in the box.
-------- the<> | in<> | --------
This is because the program finds the words print, text and file as the first order co-occurrences (as explained in case a) and finds bigrams
print<>in<> text<>the<> file<>the<> file<>in
where 'the' and 'in' co-occur with the words print, text, file.
(c) To see how the 2nd order co-occurrences of word 'line' are reached run the program using the following command - kocos.pl --order 2 --trace test.trace test.input line
and a detailed report of co-occurrence chains in test.trace file as shown in the box below.
test.trace => ---------------- line->text->the| line->file->the| line->file->in | line->print->in| ----------------
where the first line shows that the word 'line' co-occurred with 'text' which co-occurred with 'the'. Hence 'the' became a 2nd order co-occurrence. Similarly, 'line' co-occurred with 'file' which in turn co-occurred with 'the' and 'in' which are therefore the 2nd order co-occurrences of 'line'.
[Niwa&Nitta94] Y. Niwa and Y. Nitta. Co-occurrence vectors from corpora vs. distance vectors from dictionaries. COLING-1994.
[Schutze98] H. Schutze. Automatic word sense discrimination. Computational Linguistics,24(1):97-123,1998.
Amruta Purandare, pura0010@umn.edu Ted Pedersen, tpederse@umn.edu Last updated on 12/05/2003 by TDP
This work has been partially supported by a National Science Foundation Faculty Early CAREER Development award (#0092784).
http://www.d.umn.edu/~tpederse/nsp.html
Copyright (C) 2002-2003, Amruta Purandare and Ted Pedersen
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to
The Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
Amruta Purandare, University of Minnesota, Duluth, pura0010@d.umn.edu Ted Pedersen, University of Minnesota, Duluth, tpederse@umn.edu
Copyright (C) 2002-2003, Amruta Purandare & Ted Pedersen
To install Text::NSP, copy and paste the appropriate command in to your terminal.
cpanm
cpanm Text::NSP
CPAN shell
perl -MCPAN -e shell install Text::NSP
For more information on module installation, please visit the detailed CPAN module installation guide.