TMUNSW: Disorder Concept Recognition and Normalization in Clinical Notes for SemEval-2014 Task 7

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 663–667, Dublin, Ireland, August 23-24, 2014. TMUNSW: Disorder Concept Recognition and Normalization in Clinical Notes for SemEval-2014 Task 7 Jitendra Jonnagaddala Translational Cancer Research Network, University of New South Wales, Sydney 2031, Australia z3339253@unsw. edu.au Manish Kumar Krishagni Solutions Pty Ltd, Armadale 6112, Australia manish.ku- mar@krishagni .com Hong-Jie Dai* Enny Rachmani Chien-Yeh Hsu Graduate Institute of Biomedical Informatics, College of Medical Science and Technology Taipei Medical University, Taipei City 110, Taiwan {hjdai, d610101005, cyhsu}@tmu.edu.tw Abstract We present our participation in Task 7 of SemEval shared task 2014. The goal of this particular task includes the identifica- tion of disorder named entities and the mapping of each disorder to a unique Uni- fied Medical Language System concept identifier, which were referred to as Task A and Task B respectively. We partici- pated in both of these subtasks and used YTEX as a baseline system. We further developed a supervised linear chain Con- ditional Random Field model based on sets of features to predict disorder men- tions. To take benefit of results from both systems we merged these results. Under strict condition our best run evaluated at 0.549 F-measure for Task A and an accu- racy of 0.489 for Task B on test dataset. Based on our error analysis we conclude that recall of our system can be signifi- cantly increased by adding more features to the Conditional Random Field model and by using another type of tag represen- tation or frame matching algorithm to deal with the disjoint entity mentions. 1 Introduction Clinical notes are rich sources of valuable pa- tient’s information. These clinical notes are often plain text records containing important entity mentions such as clinical findings, procedures and disease mentions (Jimeno et al., 2008). Using au- tomated tools to extract the aforementioned infor- mation can undoubtedly help researchers and cli- nicians with better decision making. An important subtask of information extraction called named entity recognition (NER) can recognize the boundary of named entity mention and classify it into a certain semantic group. The focus of the SemEval-2104 task 7 is recogni- tion and normalization of disorder entities men- tioned in clinical notes. As such, this task was fur- ther divided into two parts: first, task A which in- cludes recognition of mention of concepts that be- long to UMLS (Unified Medical Language Sys- tem) semantic group disorders (Bodenreider, 2004). The concepts considered in Task A include the following eleven UMLS semantic types: Con- genital Abnormality; Acquired Abnormality; In- jury or Poisoning; Pathologic Function; Disease or Syndrome; Mental or Behavioral Dysfunction; Cell or Molecular Dysfunction; Experimental Model of Disease; Anatomical Abnormality; Ne- oplastic Process; and Signs and Symptoms. Sec- ond, task B referred to as task of normalization in- volves the mapping of each disorder mention to a UMLS concept unique identifier (CUI).The map- ping was limited to UMLS CUI of SNOMED clin- ical term codes (Spackman, Campbell, & CÃ, 1997). We participated in both tasks and devel- …

Content truncated to 500 words