Glad you enjoyed it! Unfortunately a paper whose takeaway message is "existing 
simple models work well on this new task" is not as easy to get published (and 
usually does not make as big a splash) as the "cool complex model obtains 0.5 
F1 improvement" paper.


Tim


________________________________
From: buddha <[email protected]>
Sent: Wednesday, July 6, 2016 8:00 PM
To: [email protected]
Subject: Re: update on temporal relations

Hi Tim,

I live for papers that include a variant of your phrase "First, it is very 
interesting that the best per- forming systems are the simplest and fastest. 
Despite the theoretical advantages of the con- ditional random field's global 
sequence opti- mization, the BIO approaches using local clas- sifiers typically 
obtain the best performance."

 I really enjoyed the paper!

~~~~~
May All Your Sequences Converge

On Jul 6, 2016, at 6:35 AM, Miller, Timothy 
<[email protected]<mailto:[email protected]>>
 wrote:

Sean, for temporal expressions specifically, see this paper from BioNLP
last year:

http://aclweb.org/anthology/W/W15/W15-3809.pdf<https://urldefense.proofpoint.com/v2/url?u=http-3A__aclweb.org_anthology_W_W15_W15-2D3809.pdf&d=CwMFaQ&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=ynYoryhR7Jc3mvPFCGBAp1o0JUPGbchTFnPifSLa1kQ&s=Bc2CRWQJd5P4XJZ1j7u9m1_maRY1OdPysu9HFMBHmiw&e=>
Timothy Miller; Steven Bethard; Dmitriy Dligach; Chen Lin; Guergana
Savova
Extracting Time Expressions from Clinical Text

in short, yes the SVM-based BIO taggers showed better performance than
the CRF or constituency-tree classifiers.

Tim


On Tue, 2016-07-05 at 21:41 +0000, Mullane, Sean *HS wrote:
Thank you for the links. Does that mean that the SVM-based approach
from your first link has performed better than the CRF-based approach
in CRFTimeAnnotator.java?



Thanks,

Sean



From: Savova, Guergana [mailto:[email protected]]
Sent: Friday, June 24, 2016 6:24 PM
To: '[email protected]'
Subject: RE: update on temporal relations




The best performing methods thus far have been released. They are
described in http://www.ncbi.nlm.nih.gov/pubmed/26521301

The results are state-of-the-art, see the recent community shared task
http://alt.qcri.org/semeval2016/task12/



We are actively working on the topic and will be releasing novel
methods as we investigate them.

Hope this helps.

--Guergana



Guergana Savova, PhD, FACMI

Associate Professor

PI Natural Language Processing Lab

Boston Children's Hospital and Harvard Medical School

300 Longwood Avenue

Mailstop: BCH3092

Enders 144.1

Boston, MA 02115

Tel: (617) 919-2972

Fax: (617) 730-0817

Harvard Scholar: http://scholar.harvard.edu/guergana_k_savova/biocv



From: Mullane, Sean *HS [mailto:[email protected]]
Sent: Friday, June 24, 2016 5:04 PM
To: '[email protected]' <[email protected]>
Subject: update on temporal relations




There was some mention previously (June 2015, I think) on the mailing
list about the next release of cTAKES including improved temporal
relation extraction. Can anyone share information about what
improvements or new features will be included and when the release is
expected?



Thanks,

Sean




Reply via email to