UArizona at the MADE1.0 NLP Challenge

Dongfang Xu; Vikas Yadav; Steven Bethard

UArizona at the MADE1.0 NLP Challenge

Dongfang Xu, Vikas Yadav, Steven Bethard ;

Proceedings of the 1st International Workshop on Medication and Adverse Drug Event Detection, PMLR 90:57-65, 2018.

Abstract

MADE1.0 is a public natural language processing challenge aiming to extract medication and adverse drug events from Electronic Health Records. This work presents NER and RI systems developed by UArizona team for the MADE1.0 competition. We propose a neural NER system for medical named entity recognition using both local and context features for each individual word and a simple but effective SVM-based pairwise relation classification system for identifying relations between medical entities and attributes. Our system achieves 81.56%, 83.18%, and 59.85% F1 score in the three tasks of MADE1.0 challenge, respectively, ranked amongst the top three teams for Task 2 and 3.

Related Material

Download PDF

@InProceedings{pmlr-v90-xu18a, title = {UArizona at the MADE1.0 NLP Challenge}, author = {Dongfang Xu and Vikas Yadav and Steven Bethard}, booktitle = {Proceedings of the 1st International Workshop on Medication and Adverse Drug Event Detection}, pages = {57--65}, year = {2018}, editor = {Feifan Liu and Abhyuday Jagannatha and Hong Yu}, volume = {90}, series = {Proceedings of Machine Learning Research}, address = {}, month = {04 May}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v90/xu18a/xu18a.pdf}, url = {http://proceedings.mlr.press/v90/xu18a.html}, abstract = {MADE1.0 is a public natural language processing challenge aiming to extract medication and adverse drug events from Electronic Health Records. This work presents NER and RI systems developed by UArizona team for the MADE1.0 competition. We propose a neural NER system for medical named entity recognition using both local and context features for each individual word and a simple but effective SVM-based pairwise relation classification system for identifying relations between medical entities and attributes. Our system achieves 81.56%, 83.18%, and 59.85% F1 score in the three tasks of MADE1.0 challenge, respectively, ranked amongst the top three teams for Task 2 and 3.} }

%0 Conference Paper %T UArizona at the MADE1.0 NLP Challenge %A Dongfang Xu %A Vikas Yadav %A Steven Bethard %B Proceedings of the 1st International Workshop on Medication and Adverse Drug Event Detection %C Proceedings of Machine Learning Research %D 2018 %E Feifan Liu %E Abhyuday Jagannatha %E Hong Yu %F pmlr-v90-xu18a %I PMLR %J Proceedings of Machine Learning Research %P 57--65 %U http://proceedings.mlr.press %V 90 %W PMLR %X MADE1.0 is a public natural language processing challenge aiming to extract medication and adverse drug events from Electronic Health Records. This work presents NER and RI systems developed by UArizona team for the MADE1.0 competition. We propose a neural NER system for medical named entity recognition using both local and context features for each individual word and a simple but effective SVM-based pairwise relation classification system for identifying relations between medical entities and attributes. Our system achieves 81.56%, 83.18%, and 59.85% F1 score in the three tasks of MADE1.0 challenge, respectively, ranked amongst the top three teams for Task 2 and 3.