TEES 2.1 analyses are now available for the test set

[Jari Björne]

Dear DDIExtraction 2013 participants,

The Turku Event Extraction System (https://github.com/jbjorne/TEES) is a free and open source natural language processing system developed for the extraction of events and relations from biomedical text. It was evaluated in the DDIExtraction 2011 task, and while it has been open source since last summer, it's now been updated in version 2.1 for better compatibility with the DDIExtraction 2013 (http://www.cs.york.ac.uk/semeval-2013/task9/) and BioNLP 2013 (http://2013.bionlp-st.org/) shared tasks. Some of you may have received the notification of the 2.1 release on the BioNLP mailing list.

Future updates on TEES 2.1 relevant for the DDIExtraction 2013 will be included on the wiki page at https://github.com/jbjorne/TEES/wiki/TEES-2.1 . If you have any questions about TEES, you can also send email to jari....@utu.fi.

As running TEES is quite time-consuming, precalculated analyses have also been prepared for the DDIExtraction 2013 task. These analyses, for both the training and test sets can be downloaded from the SemEval2013 FTP server at directory /home/task9/resources. BLLIP Penn-tree parses and Stanford dependency parses are provided for all datasets. Drug entity syntactic heads and TEES-predicted drug-drug interaction candidates are provided for the training set and the task 9.2 test set.

The TEES analyses for the training set were also published earlier. If some of you have used the first version (DDI13-train-TEES-analyses-130224.xml.gz) please note that in the new test set and updated training set analyses also negative interaction candidates are included. As TEES produces an SVM confidence score for each prediction, these negative classifications may be helpful when using the data in your own systems.

It's important to note that the University of Turku is not affiliated with the SemEval-2013 workshop and that the TEES 2.1 analyses are not in any way part of the official training or test data. All of these analyses have been computed automatically and you use them entirely at your own risk. That said, hopefully they prove useful, and can help you in building the best possible DDI extraction system!

Sincerely,

Jari Björne

PhD Student

University of Turku