Turku Event Extraction System 2.2
=================================
Turku Event Extraction System (TEES) is a free and open source natural language
processing system developed for the extraction of events and relations from
biomedical text. It is written mostly in Python, and should work in generic
Unix/Linux environments.
TEES has been evaluated in the following Shared Tasks and models for predicting
their targets are included in this release.
* BioNLP 2009 Shared Task (1st place)
* BioNLP 2011 Shared Task (1st place in 4/8 tasks, only system to participate
in all tasks)
* DDI 2011 (Drug-drug interactions) Challenge (4th place, at 96% of the
performance of the best system)
For more information and documentation, see the TEES wiki at
https://github.com/jbjorne/TEES/wiki
Quick Start
===========
To get started with TEES, download the latest stable release from
http://sourceforge.net/projects/tees/files or the current
version from the repository. After downloading, TEES can optionally be installed
as a module using "setup.py", but this is not required, and the program can
simply be used from the unpacked archive.
However, before using TEES the external programs and datafiles need to be
installed using the interactive configuration tool "configure.py", located
in the package root directory:
python configure.py
After TEES had been configured, you can predict events or relations for text
with classify.py. Using the "-m" (model) switch, you can select one of the
pre-computed models (listed at https://github.com/jbjorne/TEES/wiki/Classifying).
For example, to run TEES prediction for the BioNLP 2011 GENIA development
corpus, use:
python classify.py -m GE11-devel -i GE11-devel -o OUTSTEM
where "OUTSTEM" is the output file stem. To try TEES on unannotated text, you
can give "classify.py" a PubMed citation id, such as:
python classify.py -m GE11 -i 9668063 -o OUTSTEM
TEES will download the abstract and use the
integrated preprocessing pipeline to split the text into sentences (with the
GENIA Sentence Splitter, http://www.nactem.ac.uk/y-matsu/geniass/), detect
named entities (with BANNER, http://banner.sourceforge.net/) and parse the
text (with BLLIP Parser using David McClosky's biomodel,
http://bllip.cs.brown.edu/resources.shtml and Stanford format
conversion, http://nlp.stanford.edu/software/lex-parser.shtml), after which
events are detected from the document.
Using TEES
==========
The primary user interface to TEES consists of the following programs
* classify.py - Predict events/relations with an existing model
* train.py - Train a new event/relation extraction model
* batch.py - Batch process large sets of input files
* configure.py - Install TEES models, external tools and corpora
* visualize.py - Visualize the events and parse for a sentence
For information on using these programs, see the TEES wiki at
https://github.com/jbjorne/TEES/wiki
TEES also has a number of modules that can be used as standalone executables,
including the wrappers for external tools such as parsers. A list of these
executables can be found at https://github.com/jbjorne/TEES/wiki/Programs