WordNet::Similarity::jcn - Perl module for computing semantic relatedness of word senses according to the method described by Jiang and Conrath (1997).
my $wn = WordNet::QueryData->new();
my $rel = WordNet::Similarity::jcn->new($wn);
my $value = $rel->getRelatedness("car#n#1", "bus#n#2");
($error, $errorString) = $rel->getError();
die "$errorString\n" if($error);
print "car (sense 1) <-> bus (sense 2) = $value\n";
This module computes the semantic relatedness of word senses according to the method described by Jiang and Conrath (1997). This measure is based on a combination of using edge counts in the WordNet 'is-a' hierarchy and using the information content values of the WordNet concepts, as described in the paper by Jiang and Conrath. Their measure, however, computes values that indicate the semantic distance between words (as opposed to their semantic relatedness). In this implementation of the measure we invert the value so as to obtain a measure of semantic relatedness. Other issues that arise due to this inversion (such as handling of zero values in the denominator) have been taken care of as special cases.
Computes the relatedness of two word senses using an information content scheme. See the discussion section below for detailed information on how the jcn measure calculates relatedness.
Parameters: two word senses in "word#pos#sense" format.
Returns: Unless a problem occurs, the return value is the relatedness score. If no path exists between the two word senses, then a large negative number is returned. If an error occures, then the error level is set to non-zero and an error string is created (see the description of getError()). Note: the error level will also be set to 1 and an an error string will be created if no path exists between the words.
The relatedness value returned by the jcn measure is equal to 1 / jcn_distance, where jcn_distance is equal to IC(synset1) + IC(synset2) - 2 * IC(lcs). The original metric proposed by Jiang and Conrath was this distance measure. By taking the multiplicative inverse of it, we have converted it to a measure of similarity, but by so doing, we have shifted the distribution of scores.
For example, if we have the following pairs of synsets with the given jcn distances:
synset1 synset2: 3
synset3 synset4: 4
synset5 synset6: 5
We observe that the difference in the score for synset1-synset2 and synset3-synset4 is the same as for synset3-synset4 and synset5-synset6. When we take the multiplicative inverse of them, we get:
synset1 synset2: .333
synset3 synset4: .25
synset5 synset6: .2
Now the difference between the scores for synset3-synset4 is less than the difference for synset1-synset2 and synset3-synset4. This can have negative consequences when computing correlation coefficients. It might be useful to compute relatedness as max_distance - jcn_distance, where max_distance is the maximum possible jcn distance between any two synsets. The original jcn distance can easily be determined by taking the inverse of the value returned: 1/score = 1/1/jcn_distance = jcn_distance.
There are two special cases that need to be handled carefully when computing relatedness; both of these involve the case when jcn_distance is zero.
In the first case, we have ic(synset1) = ic(synset2) = ic(lcs) = 0. In an ideal world, this would only happen when all three concepts, viz. synset1, synset2, and lcs, are the root node. However, when a synset has a frequency count of zero, we use the value 0 for the information content. In this first case, we return 0 due to lack of data.
In the second case, we have ic(synset1) + ic(synset2) = 2 * ic(lics). This is almost always found when synset1 = synset2 = lcs (i.e., the two input synsets are the same). Intuitively this is the case of maximum relatedness, which would be infinity, but it is impossible to return infinity. Insteady we find the smallest possible distance greater than zero and return the multiplicative inverse of that distance.
The semantic relatedness modules in this distribution are built as classes that define the following methods: new() getRelatedness() getError() getTraceString()
See the WordNet::Similarity(3) documentation for details of these methods.
To create an object of the jcn measure, we would have the following lines of code in the Perl program.
$measure = WordNet::Similarity::jcn->new($wn, '/home/sid/jcn.conf');
The reference of the initialized object is stored in the scalar variable '$measure'. '$wn' contains a WordNet::QueryData object that should have been created earlier in the program. The second parameter to the 'new' method is the path of the configuration file for the jcn measure. If the 'new' method is unable to create the object, '$measure' would be undefined. This, as well as any other error/warning may be tested.
die "Unable to create object.\n" if(!defined $measure);
($err, $errString) = $measure->getError();
die $errString."\n" if($err);
To find the semantic relatedness of the first sense of the noun 'car' and the second sense of the noun 'bus' using the measure, we would write the following piece of code:
$relatedness = $measure->getRelatedness('car#n#1', 'bus#n#2');
To get traces for the above computation:
However, traces must be enabled using configuration files. By default traces are turned off.
The behavior of the measures of semantic relatedness can be controlled by using configuration files. These configuration files specify how certain parameters are initialized within the object. A configuration file may be specified as a parameter during the creation of an object using the new method. The configuration files must follow a fixed format.
Every configuration file starts with the name of the module ON THE FIRST LINE of the file. For example, a configuration file for the jcn module will have on the first line 'WordNet::Similarity::jcn'. This is followed by the various parameters, each on a new line and having the form 'name::value'. The 'value' of a parameter is optional (in case of boolean parameters). In case 'value' is omitted, we would have just 'name::' on that line. Comments are supported in the configuration file. Anything following a '#' is ignored till the end of the line.
The module parses the configuration file and recognizes the following parameters:
The value of this parameter specifies the level of tracing that should be employed for generating the traces. This value is an integer equal to 0, 1, or 2. If the value is omitted, then the default value, 0, is used. A value of 0 switches tracing off. A value of 1 or 2 switches tracing on. A trace level of 1 means the synsets are represented as word#pos#sense strings, while for level 2, the synsets are represented as word#pos#offset strings.
The value of this parameter specifies whether or not caching of the relatedness values should be performed. This value is an integer equal to 0 or 1. If the value is omitted, then the default value, 1, is used. A value of 0 switches caching 'off', and a value of 1 switches caching 'on'.
The value of this parameter indicates the size of the cache, used for storing the computed relatedness value. The specified value must be a non-negative integer. If the value is omitted, then the default value, 5,000, is used. Setting maxCacheSize to zero has the same effect as setting cache to zero, but setting cache to zero is likely to be more efficient. Caching and tracing at the same time can result in excessive memory usage because the trace strings are also cached. If you intend to perform a large number of relatedness queries, then you might want to turn tracing off.
The value of this parameter indicates whether or not a unique root node should be used. In WordNet, there is no unique root node for the noun and verb taxonomies. If this parameter is set to 1 (or if the value is omitted), then certain measures (wup, path, lch, res, lin, and jcn) will "fake" a unique root node. If the value is set to 0, then no unique root node will be used. If the value is omitted, then the default value, 1, is used.
The value for this parameter should be a string that specifies the path of an information content file containing the frequency of occurrence of every WordNet concept in a large corpus. A number of utility programs are included in this distribution that can be used to generate an infocontent file (see utils.pod). If no path is specified, then the default infocontent file is used, which was generated from SemCor using the sense-tags.
perl(1), WordNet::Similarity(3), WordNet::QueryData(3)
Ted Pedersen, University of Minnesota Duluth
tpederse at d.umn.edu
Siddharth Patwardhan, University of Utah
sidd at cs.utah.edu
Jason Michelizzi, University of Minnesota Duluth
mich0212 at d.umn.edu
Copyright (c) 2005, Ted Pedersen, Siddharth Patwardhan and Jason Michelizzi
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.
Note: a copy of the GNU General Public License is available on the web at http://www.gnu.org/licenses/gpl.txt and is included in this distribution as GPL.txt.
To install WordNet::Similarity, copy and paste the appropriate command in to your terminal.
perl -MCPAN -e shell
For more information on module installation, please visit the detailed CPAN module installation guide.