BioinformaticsUA: Concept Recognition in Clinical Narratives Using a Modular and Highly Efficient Text Processing Framework

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 135–139, Dublin, Ireland, August 23-24, 2014. BioinformaticsUA: Concept Recognition in Clinical Narratives Using a Modular and Highly Efficient Text Processing Framework S´ergio Matos DETI/IEETA University of Aveiro 3810-193 Aveiro, Portugal aleixomatos@ua.pt Tiago Nunes DETI/IEETA University of Aveiro 3810-193 Aveiro, Portugal tiago.nunes@ua.pt Jos´e Lu´ıs Oliveira DETI/IEETA University of Aveiro 3810-193 Aveiro, Portugal jlo@ua.pt Abstract Clinical texts, such as discharge sum- maries or test reports, contain a valuable amount of information that, if efficiently and effectively mined, could be used to infer new knowledge, possibly leading to better diagnosis and therapeutics. With this in mind, the SemEval-2014 Analysis of Clinical Text task aimed at assessing and improving current methods for identi- fication and normalization of concepts oc- curring in clinical narrative. This paper describes our approach in this task, which was based on a fully modular architec- ture for text mining. We followed a pure dictionary-based approach, after perform- ing error analysis to refine our dictionaries. We obtained an F-measure of 69.4% in the entity recognition task, achieving the second best precision over all submitted runs (81.3%), with above average recall (60.5%). In the normalization task, we achieved a strict accuracy of 53.1% and a relaxed accuracy of 87.0%. 1 Introduction Named entity recognition (NER) is an information extraction task where the aim is to identify men- tions of specific types of entities in text. This task has been one of the main focus in the biomedi- cal text mining research field, specially when ap- plied to the scientific literature. Such efforts have led to the development of various tools for the recognition of diverse entities, including species names, genes and proteins, chemicals and drugs, anatomical concepts and diseases. These tools use This work is licensed under a Creative Commons Attribution 4.0 International Licence. Page numbers and proceedings footer are added by the organisers. Licence details: http: //creativecommons.org/licenses/by/4.0/ methods based on dictionaries, rules, and machine learning, or a combination of those depending on the specificities and requirements of each concept type (Campos et al., 2013b). After identifying en- tities occurring in texts, it is also relevant to dis- ambiguate those entities and associate each occur- rence to a specific concept, using an univocal iden- tifier from a reference database such as Uniprot1 for proteins, or OMIM2 for genetic disorders. This is usually performed by matching the identified entities against a knowledge-base, possibly eval- uating the textual context in which the entity oc- curred to identify the best matching concept. The SemEval-2014 Analysis of Clinical Text task aimed at the identification and normalization of concepts in clinical narrative. Two subtasks were defined, where Task A was focused on the recognition of entities belonging to the ‘disorders’ semantic group of the Unified Medical Language System (UMLS), and Task B was focused on nor- malization of these entities to a specific UMLS Concept Unique Identifier (CUI). Specifically, the task definition required that concepts should only be normalized to CUIs that could be mapped …

Content truncated to 500 words