UMLS::Similarity
SYNOPSIS
This package consists of Perl modules along with supporting Perl
programs that implement the semantic similarity and relatedness
measures described by Leacock & Chodorow (1998), Wu & Palmer (1994),
Nguyen and Al-Mubaid (2006), Rada, et. al. 1989, Jiang & Conrath
(1997), Resnik (1995), Lin (1998), Banerjee and Pedersen(2002),
Patwardhan and Pedersen (2006) and a simple path based measure.
UMLS::Similarity requires the UMLS::Interface module to access
the Unified Medical Language System (UMLS) in order to determine
the similarity between two UMLS concepts.
The Perl modules are designed as objects with methods that take as
input two concepts from the UMLS. The semantic relatedness of these
concepts is returned by these methods. A quantitative measure of
the degree to which the two concepts are related has wide ranging
applications in numerous areas, such as word sense disambiguation,
information retrieval, etc. For example, in order to determine which
sense of a given word is being used in a particular context, the sense
having the highest relatedness with its context word senses is most
likely to be the sense being used. Similarly, in information retrieval,
retrieving documents containing highly related concepts are more likely
to have higher precision and recall values.
The following sections describe the organization of this software
package and how to use it. A few typical examples are given to help
clearly understand the usage of the modules and the supporting
utilities.
INSTALL
To install these modules run:
perl Makefile.PL
make
make test
make install
This will install the modules in the standard locations. You will,
most probably, require root privileges to install in standard system
directories. To install in a non-standard directory, specify a prefix
during the 'perl Makefile.PL' stage as:
perl Makefile.PL PREFIX=/home
It is possible to modify other parameters during installation. The
details of these can be found in the ExtUtils::MakeMaker
documentation. However, it is highly recommended not messing
around with other parameters, unless you know what you're doing.
SEMANTIC RELATEDNESS
We observe that humans find it extremely easy to say if two words are
related and if one word is more related to a given word than another.
For example, if we come across two words -- 'car' and 'bicycle', we know
they are related as both are means of transport. Also, we easily observe
that 'bicycle' is more related to 'car' than 'fork' is. But is there
some way to assign a quantitative value to this relatedness? Some ideas
have been put forth by researchers to quantify the concept of
relatedness of words, with encouraging results.
A number of different measures of relatedness have been implemented in
this software package. These include a simple edge counting
approach. The measures require the UMLS-Interface that define UMLS
concepts, and some basic relationships between these concepts.
CONTENTS
All the modules that will be installed in the Perl system directory are
present in the '/lib' directory tree of the package. These include the
semantic relatedness modules --
UMLS/Similarity/lch.pm
UMLS/Similarity/path.pm
UMLS/Similarity/wup.pm
UMLS/Similarity/nam.pm
UMLS/Similarity/cdist.pm
UMLS/Similarity/res.pm
UMLS/Similarity/lin.pm
UMLS/Similarity/jcn.pm
UMLS/Similarity/random.pm
UMLS/Similarity/vector.pm (beta)
UMLS/Similarity/lesk.pm (beta)
-- present in the lib/ subdirectory. All these modules, once installed
in the Perl system directory, can be directly used by Perl programs.
The package contains a utils/ directory that contain Perl utility
programs. These utilities use the modules or provide some supporting
functionality.
umls-similarity.pl -- returns the semantic similarity of two
terms or UMLS CUIs given a specified
measure (and view of the UMLS).
spearman.pl -- calculates the Spearman Rank
Correlation between two files
vector-input.pl -- creates the matrix and index files
required for the vector measure
SignificanceTesting.r -- R script to calculate the correlation
between a gold standard and the results
obtained using the measures in the
umls-similarity.pl program
sim2r.pl -- converts umls-similarity.pl output to
a format that can be read by the R script
create-icfrequency.pl -- create the frequency file required for
information content measures
create-icpropagation.pl -- create the probability file required for
information content measures
The package contains a web/ directory which contains a web interface to
the UMLS-Similarity package once it is installed. Please see the README
file in the web/ directory for further information.
CONFIGURATION
UMLS-Interface allows information to be extracted from the UMLS given a
specified set of sources and relations through the use of a
configuration file.
There are six configuration options: SAB, REL, RELA, SABDEF, RELDEF, and
RELADEF.
The SAB and REL options are used to determine which sources and
relations the path information is to be obtained from. The RELA option
narrows down the relation even further. The RELA will only be applied to
the PAR/CHD and RB/RN relations.
The SABDEF and RELDEF options are used to determine which sources and
relations to use when creating the Extended Definition. The RELA option
narrows down the relation even further. The RELADEF will only be applied
to the PAR/CHD and RB/RN relations.
The path, wup, lch, lin, jcn and res measures require the SAB and REL
options to be set. There is also an optional RELA option.
The vector and lesk measures require the SABDEF and RELDEF options to be
set with an optional RELADEF.
You can specify a single source, multiple sources or the entire UMLS
(using the UMLS_ALL option). Keep in mind that the greater the number of
sources the larger the search space so if you obtaining path information
about two concepts this will take longer. The names of the sources in
the configuration file are expected to be in the SAB (source
abbreviation) form. A listing of the sources and their SABs can be
found:
<http://www.nlm.nih.gov/research/umls/knowledge_sources/metathesaurus/re
lease/source_vocabularies.html>
You can specify any relations that exist in the specified set of sources
that you defined. The directional (hierarchical) relations though are
PAR/CHD and RB/RN. The other relations (such as RO and SIB) are not
directional which means when obtaining path information when using these
relations may take much longer than obtaining path information using the
directional relations. A listing of the different relations can be found
here (scroll down to the REL table):
<http://www.nlm.nih.gov/research/umls/knowledge_sources/metathesaurus/re
lease/abbreviations.html>
If you do plan on using a multiple sources or the entire UMLS, we would
advise you to use the --realtime option which is explained below, in the
Interface.pm documentation and the path programs in the utils/
directory. We also have a am UMLS_ALL option for this so you do not have
to specify each and every source and relation.
The format of the configuration file is as follows:
SAB :: <include|exclude> <source1, source2, ... sourceN>
REL :: <include|exclude> <relation1, relation2, ... relationN>
RELA :: <include|exclude> <rela1, rela2, ... relaN>
For example, if we wanted to use the MSH vocabulary with only the RB/RN
relations, the configuration file would be:
SAB :: include MSH
REL :: include RB, RN
or
SAB :: include MSH
REL :: exclude PAR, CHD
If we wanted to use the SNOMEDCT vocabulary with only the PAR/CHD
relations that are is-a relations, the configuration file would be:
SAB :: include SNOMEDCT
REL :: include PAR, CHD
RELA :: include isa, inverse_isa
The format for SABDEF and RELDEF is similar.
The SABDEF and RELDEF options are used to determine the sources and
relations the extended definition is to be obtained from.
The format of the configuration file is as follows:
SABDEF :: <include|exclude> <source1, source2, ... sourceN>
RELDEF :: <include|exclude> <relation1, relation2, ... relationN>
RELADEF :: <include|exclude> <rela1, rela2, ... relaN>
Note: RELDEF takes any of MRREL relations and two special 'relations':
1. CUI which refers to the CUIs definition
2. TERM which refers to the terms associated with the CUI
For example, if we wanted to use the definitions from MSH vocabulary and
we only wanted the definition of the CUI and the definitions of the CUIs
SIB relation, the configuration file would be:
SABDEF :: include MSH
RELDEF :: include CUI, SIB
If you wanted only the PAR/CHD definitions which are is-a relations.
SABDEF :: include MSH
RELDEF :: include PAR, CHD
RELADEF :: include isa, inverse_isa
For all of these options, there is an UMLS_ALL tag. If used with SAB or
SABDEF, it would include all of the UMLS sources. If used with the REL
or RELDEF, it would include all of the possible relations (as well as
CUI and TERM for RELDEF). If used with the RELA or RELADEF, it would
include all of the RELA relations including those with no RELA relation.
Note that this is also the default for this option which is why it is
optional. An example of using the UMLS_ALL option is as follows:
SAB :: include UMLS_ALL
REL :: include UMLS_ALL
and another is:
SABDEF :: include UMLS_ALL
RELDEF :: include UMLS_ALL
If you go to the configuration file directory, there will be example
configuration files for the different runs that you have performed.
For more information about the configuration options please see the
README.
PROPAGATION
The Information Content (IC) is defined as the negative log of the
probability of a concept. The probability of a concept, c, is determine
by summing the probability of the concept occurring in some text plus
the probability its descendants occurring in some text.
The following is an example of the method UMLS-Interface uses to
propagation counts to determine the probability of a concept in the
sources/relations specified in the configuration file. In this method,
we percolate the counts up the hierarchy, and in the case of multiple
inheritance, we send a full count up all the paths to the parent.
The icfrequency file contains the frequency of the following concepts
existing in some corpus. For example, our corpus consists of three
concepts, A B & C, each occurring five times:
SAB :: (include|exclude) <sources>
REL :: (include|exclude) <relations>
N :: 15
A<>5
B<>5
C<>5
In this case, our sources and relations consist of the following
'graph': Notation....A->D means A is a child of D....
A->D
B->D
B->E
D->F
C->E
E->F
So A B and C are "leaf" nodes and F is the root.
Step 1: determine the descendants of each nodes
Descendants(A) = {}
Descendants(B) = {}
Descendants(C) = {}
Descendants(D) = {A, B}
Descendants(E) = {B, C}
Descendants(F) = {A, B, C, D, E, F}
Step 2: determine the probability of a concept, P(c), occurring by
summing the probability of each of descendants plus its probability.
P(A) = freq(A) / N = .33
P(B) = freq(B) / N = .33
P(C) = freq(C) / N = .33
P(D) = (freq(A)+freq(B)+freq(D)) / N = .66
P(E) = (freq(B)+freq(C)+freq(E)) / N = .66
P(F) = (freq(A)+freq(B)+freq(C)+freq(D)+freq(E)+freq(F)) / N = .99
Step 3: print the probability of the concept occurring, P(c), for each
node in the sources/relations defined in the configuration table.
SMOOTH :: 0 <- or 1 if smoothing was used
SAB :: (include|exclude) <sources>
REL :: (include|exclude) <relations>
RELA :: (include|exclude) <relas> <- if any are specified in the config
N :: 15
A<>.33
B<>.33
C<>.33
D<>.66
E<>.66
F<>.99
The information content for the nodes is then calculated by taking -log
of this probability.
We have an option that incorporates Laplace smoothing. Laplace smoothing
is where the frequency count of each of the concepts in the taxonomy is
incremented by one. The advantage of doing this is that it avoids having
a concept that has a probability of zero. The disadvantage is that it
can shift the overall probability mass of the concepts from what is
actually seen in the corpus.
SOFTWARE COPYRIGHT AND LICENSE
Copyright (C) 2004-2011 Bridget T McInnes, Siddharth Patwardhan,
Serguei Pakhomov and Ted Pedersen
This suite of programs 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: The text of the GNU General Public License is provided in the file
'GPL.txt' that you should have received with this distribution.
REFERENCING
If you write a paper that has used UMLS-Similarity in some way, we'd
certainly be grateful if you sent us a copy and referenced UMLS-Interface.
We have a published paper that provides a suitable reference:
@inproceedings{McInnesPP09,
title={{UMLS-Interface and UMLS-Similarity : Open Source
Software for Measuring Paths and Semantic Similarity}},
author={McInnes, B.T. and Pedersen, T. and Pakhomov, S.V.},
booktitle={Proceedings of the American Medical Informatics
Association (AMIA) Symposium},
year={2009},
month={November},
address={San Fransisco, CA}
}
This paper is also found in
<http://www-users.cs.umn.edu/~bthomson/publications/pubs.html>
or
<http://www.d.umn.edu/~tpederse/Pubs/amia09.pdf>
REFERENCES
1 Wu Z. and Palmer M. 1994. Verb Semantics and Lexical Selection. In
Proceedings of the 32nd Annual Meeting of the Association for
Computational Linguistics. Las Cruces, New Mexico.
2 Resnik P. 1995. Using information content to evaluate semantic
similarity. In Proceedings of the 14th International Joint
Conference on Artificial Intelligence, pages 448-453, Montreal.
3 Jiang J. and Conrath D. 1997. Semantic similarity based on corpus
statistics and lexical taxonomy. In Proceedings of International
Conference on Research in Computational Linguistics, Taiwan.
4 Fellbaum C., editor. WordNet: An electronic lexical database. MIT
Press, 1998.
5 Leacock C. and Chodorow M. 1998. Combining local context and WordNet
similarity for word sense identification. In Fellbaum 1998, pp.
265-283.
6 Lin D. 1998. An information-theoretic definition of similarity. In
Proceedings of the 15th International Conference on Machine
Learning, Madison, WI.
7 Hirst G. and St-Onge D. 1998. Lexical Chains as representations of
context for the detection and correction of malapropisms. In
Fellbaum 1998, pp. 305-332.
8 Schütze H. 1998. Automatic Word Sense Discrimination. Computational
Linguistics, 24(1):97-123.
9 Resnik P. 1999. Semantic Similarity in a Taxonomy: An Information-
Based Measure and its Applications to Problems of Ambiguity in
Natural Language. Journal of Artificial Intelligence Research, 11,
95-130.
10 Budanitsky A. and Hirst G. 2001. Semantic distance in WordNet: An
experimental, application-oriented evaluation of five measures. In
Workshop on WordNet and Other Lexical Resources, Second meeting of
the North American Chapter of the Association for Computational
Linguistics. Pittsburgh, PA.
11 Banerjee S. and Pedersen T. 2002. An Adapted Lesk Algorithm for Word
Sense Disambiguation Using WordNet. In Proceeding of the Fourth
International Conference on Computational Linguistics and
Intelligent Text Processing (CICLING-02). Mexico City.
12 Patwardhan S., Banerjee S. and Pedersen T. 2002. Using Semantic
Relatedness for Word Sense Disambiguation. In Proceedings of the
Fourth International Conference on Intelligent Text Processing and
Computational Linguistics, Mexico City.
13 Banerjee S. Adapting the Lesk algorithm for word sense
disambiguation to WordNet. Master Thesis, University of Minnesota,
Duluth, 2002.
14 Patwardhan S. Incorporating dictionary and corpus information into a
vector measure of semantic relatedness. Master Thesis, University of
Minnesota, Duluth, 2003.
15 Patwardhan, S. and Pedersen T. Using WordNet Based Context Vectors
to Estimate the Semantic Relatedness of Concepts. In Proceedings of
the EACL 2006 Workshop Making Sense of Sense - Bringing Computational
Linguistics and Psycholinguistics Together, pp. 1-8, April 4, 2006,
Trento, Italy.
16 Rada, R., Mili, H., Bicknell, E. and Blettner, M. Development and
application of a metric on semantic nets. In Proceedings of the
IEEE Transactions on Systems, Man, and Cybernetics, volume 19,
pages 17-30, 1989.
17 Nguyen, H.A. and Al-Mubaid, H. New ontology based semantic
similarity mesaure for the biomedical domain. In Proceedings of
the IEEE International Conference on Granular Computing, pages
623-628, 2006.
SEE ALSO
<http://search.cpan.org/dist/UMLS-Interface>
<http://search.cpan.org/dist/UMLS-Similarity>
CONTACT US
If you have any trouble installing and using UMLS-Interface, please
contact us via the users mailing list :
umls-similarity@yahoogroups.com
You can join this group by going to:
<http://tech.groups.yahoo.com/group/umls-similarity/>
You may also contact us directly if you prefer :
Bridget T. McInnes: bthomson at cs.umn.edu
Ted Pedersen : tpederse at d.umn.edu
AUTHORS
Bridget T McInnes, University of Minnesota Twin Cities
bthomson at cs.umn.edu
Siddharth Patwardhan, University of Utah
sidd at cs.utah.edu
Serguei Pakhomov, University of Minnesota Twin Cities
pakh002 at umn.edu
Ted Pedersen, University of Minnesota Duluth
tpederse at d.umn.edu
Ying Liu, University of Minnesota
liux0395 at umn.edu
DOCUMENTATION COPYRIGHT AND LICENSE
Copyright (C) 2003-2011 Bridget T. McInnes, Siddharth Patwardhan,
Serguei Pakhomov, Ying Liu and Ted Pedersen.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no
Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts.
Note: a copy of the GNU Free Documentation License is available on the
web at:
<http://www.gnu.org/copyleft/fdl.html>
and is included in this distribution as FDL.txt.