eprintid: 2472
rev_number: 3
eprint_status: archive
userid: 307
dir: disk0/00/00/24/72
datestamp: 2017-06-10 11:32:27
lastmod: 2017-06-10 11:32:27
status_changed: 2017-06-10 11:32:27
type: article
succeeds: 2001
metadata_visibility: show
creators_name: Le, Hoang Quynh
creators_name: Tran, Mai Vu
creators_name: Dang, Thanh Hai
creators_name: Ha, Quang Thuy
creators_name: Collier, Nigel
creators_id: lhquynh@gmail.com
creators_id: vutm@vnu.edu.vn
creators_id: hai.dang@vnu.edu.vn
creators_id: thuyhq@vnu.edu.vn
title: Sieve-based coreference resolution enhances semi-supervised learning model for chemical-induced disease relation extraction
ispublished: pub
subjects: IT
subjects: isi
divisions: fac_fit
abstract: The BioCreative V chemical-disease relation (CDR) track was proposed to accelerate progress of text mining in facilitating integrative understanding ofchemical substances, diseases and their relations. In this article, we describe an extension of the UET-CAM system for mining chemical-disease relations from text data, of which performance was ranked 4th among 18 participating corresponding systems by the BioCreative CDR track committee. In Disease Named Entity Recognition and Normalization (DNER) phase, our system employs joint learning with a perceptron-based named entity recognizer (NER) and a back-off model with Semantic Supervised Indexing (SSI) and Skip-gram for named entity  normalization  (NEN).  Crucially,  for  solving  the  chemical-induced  disease  (CID)  sub-task,  we propose a pipeline that includes a coreference resolution module and a SVM intra-sentence relations extraction model. The former module utilizes a multi-pass sieve to identify inter-sentence references for entities while the latter is trained on both the CDR data and our silverCID corpus with a rich feature set. SilverCID is the silver standard corpus contains more than 50 thousands sentences which are automatically built based on the CTD database in order to provide evidence for the CID relation extraction. We critically evaluated our method on the CDR test set in order to clarify the contribution of our system components. Results show an F1 of 82.44 for the DNER task, and a best performance of F1 58.90 on the CID task. The comparisons also demonstrate the significant contribution of the  multi-pass sieve coreference resolution method and the silverCID corpus.
date: 2016
date_type: published
official_url: http://doi.org/10.1093/database/baw102
id_number: doi:10.1093/database/baw102
full_text_status: public
publication: Database
volume: 2016
pagerange: baw102
refereed: TRUE
issn: 1758-0463
related_url_url: http://database.oxfordjournals.org/content/2016/baw102
citation:   Le, Hoang Quynh and Tran, Mai Vu and Dang, Thanh Hai and Ha, Quang Thuy and Collier, Nigel  (2016) Sieve-based coreference resolution enhances semi-supervised learning model for chemical-induced disease relation extraction.  Database, 2016 .   baw102.  ISSN 1758-0463     
document_url: https://eprints.uet.vnu.edu.vn/eprints/id/eprint/2472/1/Sieve-based%20coreference%20resolution%20enhances%20semi-supervised%20learning%20model%20for%20chemical-induced%20disease%20relations%20extraction.pdf