BluLab: Temporal Information Extraction for the 2015 Clinical TempEval Challenge

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 815–819, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics BluLab: Temporal Information Extraction for the 2015 Clinical TempEval Challenge Sumithra Velupillai1,2, Danielle L Mowery2, Samir Abdelrahman2, Lee Christensen2 and Wendy W Chapman2 1Dept. of Computer and Systems Sciences (DSV) Stockholm University, Sweden 2Department of Biomedical Informatics University of Utah, Salt Lake City sumithra@dsv.su.se, {firstname.lastname}@utah.edu Abstract The 2015 Clinical TempEval Challenge ad- dressed the problem of temporal reasoning in the clinical domain by providing an anno- tated corpus of pathology and clinical notes related to colon cancer patients. The chal- lenge consisted of six subtasks: TIMEX3 and event span detection, TIMEX3 and event at- tribute classification, document relation time and narrative container relation classification. Our BluLab team participated in all six sub- tasks. For the TIMEX3 and event subtasks, we developed a ClearTK support vector ma- chine pipeline using mainly simple lexical fea- tures along with information from rule-based systems. For the relation subtasks, we em- ployed a conditional random fields classifica- tion approach, with input from a rule-based system for the narrative container relation sub- task. Our team ranked first for all TIMEX3 and event subtasks, as well as for the docu- ment relation subtask. 1 Introduction Temporal information extraction plays a crucial role in improved information access, in particular for creating timelines and detailed question answering. Several previous natural language processing (NLP) research community challenges have dealt with tem- poral reasoning in the newswire domain (Verhagen et al., 2010; UzZaman et al., 2013) and the clinical domain (Sun et al., 2013). The 2015 Clinical TempEval challenge (Bethard et al., 2015) addressed temporal reasoning subtasks similar to these previous efforts by providing a new benchmark corpus in the clinical domain with anno- tated pathology and clinical notes from colon can- cer patients. The corpus is annotated with a modi- fied version of the TimeML schema (Pustejovsky et al., 2010), where adaptations specific to this domain have been developed (Styler et al., 2014). For successful temporal modelling, three core concepts need to be defined: temporal expressions (TIMEX3), denoting time references like dates; events (EVENT), denoting salient occurrences; and temporal relations (TLINK) denoting order (e.g. before, after) between an event and/or TIMEX3. As part of the 2012 i2b2/VA Challenge, the best performing systems for classification of TIMEX3 (F1: 0.66), EVENTS (F1: 0.92), their attributes (av- erage accuracy: 0.86) and TLINKS (F1: 0.69) ap- plied regular expressions as well as machine learn- ing approaches such as conditional random fields (CRF) and support vector machines (SVM) (Sun et al., 2013). For the 2013/2014 CLEF/ShARe Chal- lenges, the best approaches for strict information extraction (F1: detection and accuracy: normaliza- tion) of TIMEXs (0.287 F1 and 0.354 accuracy), disease/disorder EVENTS (0.750 F1 and 0.589 ac- curacy), and EVENT attributes (0.676 F1 and 0.868 accuracy) leveraged the Apache cTAKES (Savova et al., 2010) framework, Begin-Inside-Outside (BIO) tagging, and CRF and SVM for (Pradhan et al., 2015; Mowery et al., 2014). The 2015 Clinical …

Content truncated to 500 words