BioinformaticsUA: Machine Learning and Rule-Based Recognition of Disorders and Clinical Attributes from Patient Notes

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 422–426, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics BioinformaticsUA: Machine Learning and Rule-Based Recognition of Disorders and Clinical Attributes from Patient Notes S´ergio Matos DETI/IEETA University of Aveiro 3810-193 Aveiro, Portugal aleixomatos@ua.pt Jos´e Sequeira DETI/IEETA University of Aveiro 3810-193 Aveiro, Portugal sequeira@ua.pt Jos´e Lu´ıs Oliveira DETI/IEETA University of Aveiro 3810-193 Aveiro, Portugal jlo@ua.pt Abstract Natural language processing and text analy- sis methods offer the potential of uncovering hidden associations from large amounts of un- processed texts. The SemEval-2015 Analy- sis of Clinical Text task aimed at fostering re- search on the application of these methods in the clinical domain. The proposed task con- sisted of disorder identification with normal- ization to SNOMED-CT concepts, and disor- der attribute identification, or template filling. We participated in both sub-tasks, using a combination of machine-learning and rules for recognizing and normalizing disease men- tions, and rule-based methods for template filling. We achieved an F-score of 71.2% in the entity recognition and normalization task, and a slot weighted accuracy of 69.5% in the template filling task. 1 Introduction Biomedical text mining offers the promise of lever- aging the huge amounts of information available on scientific documents to help raise new hypothe- ses and uncover hidden knowledge. Biomedical text mining (TM) has been an important focus of research during the last years, sustained by the high volumes of data, the diverse computational and multi-disciplinary challenges posed, and by the potential impact of new discoveries (Simpson and Demner-Fushman, 2012). These benefits have been demonstrated in recent studies in which text mining methods were used to suggest biomarkers for diag- nosis and for measuring disease progression, targets for new drugs, or new uses for existing drugs (Fri- jters et al., 2010). Likewise, clinical information stored as natural language text in discharge notes and reports could be exploited to identify important associations, and this has led to an increased interest in applying text mining techniques to such texts, in order to extract information related to diseases, med- ications, and adverse drug events, for example (Zhu et al., 2013). Research efforts in biomedical text mining have led to the development of various methods and tools for the recognition of diverse entities, including species names, genes and proteins, chemicals and drugs, anatomical concepts and diseases. These methods are based on dictionaries, rules, and ma- chine learning, or a combination of those depend- ing on the specificities and requirements of each concept type. After identifying entity mentions in text, it becomes necessary to perform entity normal- ization, which consists in assigning a specific con- cept identifier to each entity. This is usually per- formed by matching the identified entities against a knowledge-base, possibly evaluating the textual context in which the entity occurred to identify the best matching concept. Following up on the 2014 task, in which the ob- jective was the identification and normalization of disease concepts in clinical texts (Pradhan et al., 2014), two subtasks were …

Content truncated to 500 words