TY - GEN
T1 - Clinical entity recognition using structural support vector machines with rich features
AU - Tang, Buzhou
AU - Cao, Hongxin
AU - Wu, Yonghui
AU - Jiang, Min
AU - Xu, Hua
PY - 2012
Y1 - 2012
N2 - Named entity recognition (NER) is an important task for natural language processing (NLP) of clinical text. Conditional Random Fields (CRFs), a sequential labeling algorithm, and Support Vector Machines (SVMs), which is based on large margin theory, are two typical machine learning algorithms that have been widely applied to NER tasks, including clinical entity recognition. However, Structural Support Vector Machines (SSVMs), an algorithm that combines the advantages of both CRFs and SVMs, has not been investigated for clinical text processing. In this study, we applied the SSVMs algorithm to the Concept Extraction task of the 2010 i2b2 clinical NLP challenge, which was to recognize entities of medical problems, treatments, and tests from hospital discharge summaries. Using the same training (N = 27,837) and test (N = 45,009) sets in the challenge, our evaluation showed that the SSVMs-based NER system required less training time, while achieved better performance than the CRFs-based system for clinical entity recognition, when same features were used. Our study also demonstrated that rich features such as unsupervised word representations improved the performance of clinical entity recognition. When rich features were integrated with SSVMs, our system achieved a highest F-measure of 85.74% on the test set of 2010 i2b2 NLP challenge, which outperformed the best system reported in the challenge by 0.5%.
AB - Named entity recognition (NER) is an important task for natural language processing (NLP) of clinical text. Conditional Random Fields (CRFs), a sequential labeling algorithm, and Support Vector Machines (SVMs), which is based on large margin theory, are two typical machine learning algorithms that have been widely applied to NER tasks, including clinical entity recognition. However, Structural Support Vector Machines (SSVMs), an algorithm that combines the advantages of both CRFs and SVMs, has not been investigated for clinical text processing. In this study, we applied the SSVMs algorithm to the Concept Extraction task of the 2010 i2b2 clinical NLP challenge, which was to recognize entities of medical problems, treatments, and tests from hospital discharge summaries. Using the same training (N = 27,837) and test (N = 45,009) sets in the challenge, our evaluation showed that the SSVMs-based NER system required less training time, while achieved better performance than the CRFs-based system for clinical entity recognition, when same features were used. Our study also demonstrated that rich features such as unsupervised word representations improved the performance of clinical entity recognition. When rich features were integrated with SSVMs, our system achieved a highest F-measure of 85.74% on the test set of 2010 i2b2 NLP challenge, which outperformed the best system reported in the challenge by 0.5%.
KW - Conditional random fields
KW - Named entity recognition
KW - Natural language processing
KW - Structural support vector machines
KW - Support vector machines
UR - https://www.scopus.com/pages/publications/84870518953
U2 - 10.1145/2390068.2390073
DO - 10.1145/2390068.2390073
M3 - 会议稿件
AN - SCOPUS:84870518953
SN - 9781450317160
T3 - International Conference on Information and Knowledge Management, Proceedings
SP - 13
EP - 19
BT - DTMBIO'12 - Proceedings of the 6th ACM International Workshop on Data and Text Mining in Biomedical Informatics, Co-located with CIKM 2012
T2 - 6th ACM International Workshop on Data and Text Mining in Biomedical Informatics, DTMBIO 2012, in Conjunction with the 21st ACM International Conference on Information and Knowledge Management, CIKM 2012
Y2 - 29 October 2012 through 29 October 2012
ER -