GUIR at SemEval-2017 Task 12: A Framework for Cross-Domain Clinical Temporal Information Extraction

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 1024–1029, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics GUIR at SemEval-2017 Task 12: A Framework for Cross-Domain Clinical Temporal Information Extraction Sean MacAvaney, Arman Cohan, and Nazli Goharian Information Retrieval Lab Department of Computer Science Georgetown University {firstname}@ir.cs.georgetown.edu Abstract Clinical TempEval 2017 (SemEval 2017 Task 12) addresses the task of cross- domain temporal extraction from clinical text. We present a system for this task that uses supervised learning for the ex- traction of temporal expression and event spans with corresponding attributes and narrative container relations. Approaches include conditional random fields and de- cision tree ensembles, using lexical, syn- tactic, semantic, distributional, and rule- based features. Our system received best or second best scores in TIMEX3 span, EVENT span, and CONTAINS relation ex- traction. 1 Introduction Clinical TempEval 2017 (Bethard et al., 2017) is designed to address the challenge of extracting clinical timelines from medical narratives. It is a successor to Clinical TempEval 2016 (Bethard et al., 2016), Clinical TempEval 2015 (Bethard et al., 2015), and the i2b2 temporal challenge (Sun et al., 2013). Clinical TempEval evaluates systems using the THYME corpus (Styler IV et al., 2014), which is annotated with temporal expressions (TIMEX3), events (EVENT), and temporal relations (TLINK) per an extension of the TimeML specifications (Pustejovsky et al., 2003). The focus of Clinical TempEval 2017 is domain adaptation. The source domain consists of clinical text about patients undergoing colon cancer treat- ments, while the target domain consists of clini- cal text about those with brain cancer. There are two phases in the task. In phase 1, the shared task provides no annotations for the target domain (un- supervised). In phase 2, the shared task provides a small annotated training set from the target do- main (supervised). Both phases evaluate system performance on thirteen tasks via precision, recall, and F1-score. In Clinical TempEval 2016, the top-performing system employed structural support vector ma- chines (SVM) for entity span extraction and lin- ear support vector machines for attribute and re- lation extraction (Lee et al., 2016). For the pre- vious iteration, Velupillai et al. (2015) developed a pipeline based on ClearTK and SVM with lex- ical and rule-based features to extract TIMEX3 and EVENT mentions. In the i2b2 2012 temporal challenge, all top performing teams used a combi- nation of supervised classification and rule-based methods for extracting temporal information and relations (Sun et al., 2013). Other efforts in clini- cal temporal annotation include works by Roberts et al. (2008), Savova et al. (2009), and Galescu and Blaylock (2012). Previous work has also investigated extract- ing temporal relations. Examples of these ef- forts in the general domain include: classification by SVM (Chambers et al., 2007), Integer Linear Programming (ILP) for temporal ordering (Cham- bers and Jurafsky, 2008), Markov Logic Networks (Yoshikawa et al., 2009), and SVM with Tree Ker- nels (Miller et al., 2013). In this paper, we present a framework for tem- poral information extraction in …

Content truncated to 500 words