A Study of Concept-based Weighting Regularization for Medical Records Search

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 603–612, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics A Study of Concept-based Weighting Regularization for Medical Records Search Yue Wang, Xitong Liu, Hui Fang Department of Electrical & Computer Engineering, University of Delaware, USA {wangyue,xtliu,hfang}@udel.edu Abstract An important search task in the biomedical domain is to find medical records of pa- tients who are qualified for a clinical trial. One commonly used approach is to apply NLP tools to map terms from queries and documents to concepts and then compute the relevance scores based on the concept- based representation. However, the map- ping results are not perfect, and none of previous work studied how to deal with them in the retrieval process. In this pa- per, we focus on addressing the limitations caused by the imperfect mapping results and study how to further improve the re- trieval performance of the concept-based ranking methods. In particular, we ap- ply axiomatic approaches and propose two weighting regularization methods that ad- just the weighting based on the relations among the concepts. Experimental results show that the proposed methods are effec- tive to improve the retrieval performance, and their performances are comparable to other top-performing systems in the TREC Medical Records Track. 1 Introduction With the increasing use of electronic health records, it becomes urgent to leverage this rich information resource about patients’ health condi- tions to transform research in health and medicine. As an example, when developing a cohort for a clinical trial, researchers need to identify pa- tients matching a set of clinical criteria based on their medical records during their hospital visits (Safran et al., 2007; Friedman et al., 2010). This selection process is clearly a domain-specific re- trieval problem, which searches for relevant medi- cal records that contain useful information about their corresponding patients’ qualification to the criteria specified in a query, e.g., “female patient with breast cancer with mastectomies during ad- mission”. Intuitively, to better solve this domain-specific retrieval problem, we need to understand the re- quirements specified in a query and identify the documents satisfying these requirements based on their semantic meanings. In the past decades, significant efforts have been put on constructing biomedical knowledge bases (Aronson and Lang, 2010; Lipscomb, 2000; Corporation, 1999) and developing natural language processing (NLP) tools, such as MetaMap, to utilize the informa- tion from the knowledge bases (Aronson, 2001; McInnes et al., 2009). These efforts make it pos- sible to map free text to concepts and use these concepts to represent queries and documents. Indeed, concept-based representation is one of the commonly used approaches that leverage knowledge bases to improve the retrieval perfor- mance (Limsopatham et al., 2013d; Limsopatham et al., 2013b). The basic idea is to represent both queries and documents as “bags of concepts”, where the concepts are identified based on the in- formation from the knowledge bases. This method has been shown to be more effective than tra- ditional term-based representation in …

Content truncated to 500 words