Hitachi at SemEval-2017 Task 12: System for temporal information extraction from clinical notes

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 1005–1009, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics Hitachi at SemEval-2017 Task 12: System for temporal information extraction from clinical notes Sarath P R1, Manikandan R1, Yoshiki Niwa2 1Research and 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 devel- oped for the task of temporal informa- tion extraction from clinical narratives in the context of the 2017 Clinical TempE- val challenge. Clinical TempEval 2017 ad- dressed the problem of temporal reason- ing in the clinical domain by providing an- notated clinical notes, pathology and ra- diology reports in line with Clinical Tem- pEval challenges 2015/16, across two dif- ferent evaluation phases focusing on cross domain adaptation. Our team focused on subtasks involving extractions of tempo- ral spans and relations for which the de- veloped systems showed average F-score of 0.45 and 0.47 across the two phases of evaluations. 1 Introduction Temporal information extraction has been a widely explored topic of research interest in the field of information extraction during recent years. It is essential for improving the performance of ap- plications such as question answering, search, text classification and systems that establish timelines from clinical narratives. In this line over the years, research community challenges on clinical tem- poral information extraction have been organized; i.e., the 2012 Informatics for Integrating Biology and the Bedside (i2b2) challenge (Sun et al., 2013) the 2013/2014 CLEF/ShARe challenge (Mowery et al., 2014), and the 2015/16 Clinical TempEval challenge (Bethard et al., 2015, 2016). These chal- lenges provide annotated corpora on temporal en- tities and relations, which facilitate comparisons of multiple systems and push the state of art in the development of clinical temporal information ex- traction methodologies. The 2017 Clinical TempEval challenge is the most recent community challenge that addresses temporal information extraction from clinical notes. The challenge was in inline with 2015/16 challenge in terms of subtasks. However this year’s challenge focussed on cross domain adap- tation across two phases of evaluation. In phase one (unsupervised domain adaptation), the sys- tems were evaluated on their results for all six sub-tasks on brain cancer notes given colon can- cer notes (data of 2015/16 challenge) as inputs. In phase two (supervised domain adaptation), evalu- ation was carried out in line with phase one but a small number of annotations of brain cancer notes were also given as inputs. In this paper, we describe an end-to-end sys- tem that addresses subtasks involving extractions of temporal spans and relations. We designed the system by adapting various insights and tech- niques from previous work on temporal informa- tion extraction in the clinical domain (Sarath et al., 2016; Abdulsalam et al., 2016; Johri et al., 2014) and ensemble modelling (Dzeroski and Zenko, 2004). The rest of the paper is organised as follows: In section 2, we discuss datasets, methods and fea- ture sets used for each of the subtasks. …

Content truncated to 500 words