Hitachi at SemEval-2016 Task 12: A Hybrid Approach for Temporal Information Extraction from Clinical Notes

Proceedings of SemEval-2016, pages 1231–1236, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics Hitachi at SemEval-2016 Task 12: A Hybrid Approach for Temporal Infor- mation Extraction from Clinical Notes Sarath P R1, Manikandan R1, Yoshiki Niwa2 1Research & Development Center, Hitachi India Pvt Ltd, India 2Hitachi Ltd, Center for Exploratory Research, Japan {sarath,manikandan}@hitachi.co.in yoshiki.niwa.tx@hitachi.com Abstract This paper describes the system developed for the task of temporal information extraction from clinical narratives in the context of 2016 Clinical TempEval challenge. Clinical TempEval 2016 addressed the problem of temporal reasoning in clinical domain by providing annotated clinical notes and pathol- ogy reports similar to Clinical TempEval chal- lenge 2015. The Clinical TempEval challenge consisted of six subtasks. Hitachi team partic- ipated in two time expression based subtasks: time expression span detection (TS) and time expression attribute identification (TA) for which we developed hybrid of rule-based and machine learning based methods using Stan- ford TokensRegex framework and Stanford Named Entity Recognizer and evaluated it on the THYME corpus. Our hybrid system achieved a maximum F-score of 0.73 for iden- tification of time spans (TS) and 0.71 for identification of time attributes (TA). 1 Introduction Temporal information extraction has been a trend- ing topic of research interest in the field of infor- mation extraction. It is crucial for improvement of systems used in number of applications ranging from question answering, search engines, text classification etc. to the systems that establish timelines and explicitly ground events that occurs in clinical narratives. This work focuses on the automatic identifica- tion of time expressions from clinical texts. Time expressions are words and phrases that corre- spond to points or spans on a timeline, such as Dates, Time, Durations, Quantifiers, Set, and Pre- postExp. Table 1 shows the time expression clas ses used in this work, with examples given for each class. Time Class Example Date February 2 2010, Friday morn- ing Time 5:30 PM, 20 minutes ago Duration For the next 24 hours, nearly 2 weeks Quantifier twice, three times Prepost postoperatively, post-surgery Set twice daily, weekly Table 1: Examples of time expressions. Research work on temporal information extrac- tion has been carried out in both general NLP domain (Verhagen et al., 2007; Verhagen et al., 2010; UzZaman et al., 2013) and in clinical do- main (Raghavan et al., 2012; Sun et al., 2013; Miller et al.,2013; Bethard et al., 2015; Velupillai et al 2015). In clinical domain, temporal infor- mation extraction has seen much research interest with assignment of medical events to time bins (Raghavan et al, 2012), i2b2 (Informatics for In- tegrating Biology and the Bedside) shared task (Sun et al., 2013) with the best performing system (Xu et al., 2013) using CRF (Conditional Random Field) and SVM (Support Vector Machines) and development of a hybrid system (Kovacevic et al., 2013) that used CRF and rule-based methods. Further tasks similar to i2b2 that involve temporal reasoning were also attempted in ShARe CLEF 2013/2014 eHealth challenges (Pradhan et al., 2015) and Clinical TempEval …

Content truncated to 500 words