siqil / udaner

Unsupervised Domain Adaptation for biomedical Named Entity Recognition

Usage

To see a list of available parameters:

java -jar target/udaner-1.0-jar-with-dependencies.jar

Using the basic CRF model:

java -jar target/udaner-1.0-jar-with-dependencies.jar -source-train data/genetag_train -target-train data/fly_train -test data/fly_test

The input file should be in the format that each line is a token and its label (seperated by TAB), and sentences are seperated by a blank line (./data directory contains some examples):

Cervicovaginal	O
foetal	B-GENE
fibronectin	I-GENE
in	O
the	O
prediction	O
of	O
preterm	O
labour	O
in	O
a	O
low	O
-	O
risk	O
population	O
.	O

Varicella	B-GENE
-	I-GENE
zoster	I-GENE
virus	I-GENE
(	I-GENE
VZV	I-GENE
)	I-GENE
glycoprotein	I-GENE
gI	I-GENE
is	O
a	O
type	B-GENE
1	I-GENE
transmembrane	I-GENE
glycoprotein	I-GENE
which	O
is	O
one	O
component	O
of	O
the	O
heterodimeric	O
gE	B-GENE
:	O
gI	B-GENE
Fc	I-GENE
receptor	I-GENE
complex	O
.	O

The default method used to train the model does not perform domain adaptation. Change the method with -method:

java -jar target/udaner-1.0-jar-with-dependencies.jar -source-train data/genetag_train -target-train data/fly_train -test data/fly_test -method FEATURE_SUBSETTING

Supported methods include LIKELIHOOD (default), FEATURE_SUBSETTING, BOOTSTRAPPING, ENTROPY_REGULARIZATION.

To use SCL, just specify -scl (SVDLIBC must be installed and svd must be in the PATH. See Installation)

java -jar target/udaner-1.0-jar-with-dependencies.jar -source-train data/genetag_train -target-train data/fly_train -test data/fly_test -scl

To output the labels for an unlabeled data set, use -pred instead of -test

java -jar target/udaner-1.0-jar-with-dependencies.jar -source-train data/genetag_train -target-train data/fly_train -pred data/fly_test

The program outputs the labels directly to STDOUT. If you want to save it in a file, you can redirect STDOUT:

java -jar target/udaner-1.0-jar-with-dependencies.jar -source-train data/genetag_train -target-train data/fly_train -pred data/fly_test > fly_test_labels

To save the model in a file and use it later, use -write-model FILE and -read-model FILE:

java -jar target/udaner-1.0-jar-with-dependencies.jar -source-train data/genetag_train -target-train data/fly_train -write-model model
java -jar target/udaner-1.0-jar-with-dependencies.jar -read-model model -test data/fly_test 

Installation

There is a compiled version at ./target/udaner-1.0-jar-with-dependencies.jar. You can just use it.

You can also build from the source code. You need to install Apache Maven first. Then you can execute

mvn install

in the current directory. The executable jar should be in ./target.

If you want to use SCL, you also need to install SVDLIBC. Its source code is also in utils/svdlibc.tgz. After installation, svd should be in the PATH of your system.

References

This work was done when I was in the Zhang Lab, Beijing Institute of Genomics. The paper describing this work is in submission. Please contat the Zhang Lab for further information. Following are the main references of this work.

1. Leaman R, Gonzalez G. BANNER: an executable survey of advances in biomedical named entity recognition. Pacific Symp Biocomput. 2008;663: 652–663. Available: http://www.ncbi.nlm.nih.gov/pubmed/18229723
2. Blitzer J, Dredze M, Pereira F. Biographies, bollywood, boom-boxes and blenders: Domain adaptation for sentiment classification. ACL. 2007; 440–447. Available: http://acl.ldc.upenn.edu/P/P07/P07-1056.pdf
3. Satpal S, Sarawagi S. Domain adaptation of conditional probability models via feature subsetting. Knowl Discov Databases PKDD 2007. 2007; Available: http://link.springer.com/chapter/10.1007/978-3-540-74976-9_23
4. Wu D, Lee WS, Ye N, Chieu HL. Domain adaptive bootstrapping for named entity recognition. Proc 2009 Conf Empir Methods Nat Lang Process. 2009; 1523–1532. Available: http://dl.acm.org/citation.cfm?id=1699699
5. Lafferty JD, McCallum A, Pereira FCN. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data. Proceedings of the Eighteenth International Conference on Machine Learning. 2001. pp. 282–289. Available: http://dl.acm.org/citation.cfm?id=645530.655813
6. McCallum AK. MALLET: A Machine Learning for Language Toolkit [Internet]. 2002. Available: http://mallet.cs.umass.edu