NTU-1 at SemEval-2017 Task 12: Detection and classification of temporal events in clinical data with domain adaptation

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 1010–1013, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics NTU-1 at SemEval-2017 Task 12: Detection and classification of tem- poral events in clinical data with supervised domain adaptation Po-Yu Huang*, Hen-Hsen Huang†, Yu-Wun Wang†, Ching Huang†, Hsin-Hsi Chen† *School of Medicine Taipei Medical University Taipei, Taiwan †Department of Computer Science and Information Engineering National Taiwan University Taipei, Taiwan b101100089@tmu.edu.tw, hhhuang@nlg.csie.ntu.edu.tw, {b02902033, b02902042, hhchen}@ntu.edu.tw Abstract This study proposes a system to automati- cally analyze clinical temporal events in a fine-grained level in SemEval-2017. Sup- port vector machine (SVM) and condition- al random field (CRF) were implemented in our system for different subtasks, in- cluding detecting clinical relevant events and time expression, determining their at- tributes, and identifying their relations with each other within the document. Do- main adaptation was the main challenge this year. Unified Medical Language Sys- tem was consulted to generalize events specific to each domain. The results showed our system’s capability of domain adaptation. 1 Introduction This study proposes a system to participate in the Clinical TempEval 2017 shared task, which fo- cuses on the detection and classification of tem- poral events in clinical data. To better utilize the information in clinical data, temporal event ex- traction is fundamental in previous researches (Bethard et al., 2016). Unlike previous studies, the training and the test data are in different do- mains this year. The task is further separated into two phases: unsupervised domain adaption and supervised domain adaption. We took part in the supervised domain adaption where data of 591 records of colon cancer patients and 30 records of brain cancer patients from Mayo clinic were giv- en. Based on the THYME corpus (Styler IV et al., 2014), we propose a framework that automatical- ly analyzes clinical temporal events in a fine- grained level. Our framework identifies temporal events in unstructured text and further labels every event with its attributes. The task consists of three major subtasks. The first one is to detect clinical relevant events and time expressions in a given medical record. From the unstructured text, both the spans of time ex- pressions and the spans of event mentions are identified. The second subtask is analyzing the attributes of time expressions and event mentions. A time expression will be classified into one of six types: DATE, TIME, DURATION, QUANTIFIER, PREPOSTEXP, and SET. An event mention con- tains four properties such as type of an event, po- larity, degree and modality. Our model labels the- se four properties to every event mention. The final subtask is to determine two kinds of relations within the text. DocTimeRel is the rela- tion between the document creation time and an event mention. Four types of DocTimeRel includ- ing BEFORE, AFTER, OVERLAP, and BE- FORE-OVERLAP are annotated in THYME. In addition to DocTimeRel, our model also recog- nizes the narrative container of an event mention called TLINK in this task. There are five types of TLINK, including …

Content truncated to 500 words