A Corpus for Evidence Based Medicine Summarisation

A Corpus for Evidence Based Medicine Summarisation Diego Molla Macquarie University Sydney, Australia diego.molla-aliod@mq.edu.au Abstract In this paper we motivate the need for a corpus for the development and testing of summarisation systems for evidence- based medicine. We describe the corpus which we are currently creating, and show its applicability by evaluating several sim- ple query-based summarisation techniques using a small fragment of the corpus. 1 Introduction Current clinical guidelines urge medical prac- titioners to practise Evidence Based Medicine (EBM) when providing care for their patients (Sackett et al., 2000). EBM has been defined as “the conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients” (Sackett et al., 1996). To find and appraise the evidence the medical practitioner has access to systematic re- views available through search tools such as the Cochrane Library1 and UpToDate2. However, there is not always a systematic review that ad- dresses the specific topic at hand (Sackett et al., 2000) and then a search on the primary literature becomes necessary. The amount of documents that exist in the pri- mary literature is overwhelming. The US Na- tional Library of Medicine, for example, offers PubMed,3 a database of medical publications that comprises more than 20 million citations. search in PubMed often returns thousands of doc-A uments. With such amount of text, summarising the information becomes crucial. The tools avail- able to the medical practitioner — see e.g. the Survey by Berkowitz (2002) — typically focus on finding and ranking the relevant papers, often with 1http://www.thecochranelibrary.com/ 2http://www.uptodateonline.com 3http://www.ncbi.nlm.nih.gov/pubmed easy access to the abstracts and type of study, and sometimes with highlight of matching terms. But surprisingly little effort has been placed on sum- marising the information for easy perusal and ap- praisal by the user. In this paper we stress the lack of corpora to help research in evidence-based summarisation of clinical articles (Section 2). We present the char- acteristics of the corpus we are developing (Sec- tion 3), and we show the use of a small fragment of the corpus for the evaluation of simple summarisa- tion techniques (Section 4). 2 Where is the Corpus for Summarisation? Current summarisation systems have been devel- oped and tested by using corpora built ad-hoc and there is no common corpus readily available specifically for the task. Afantenos et al. (2005) surveys research in summarisation from medical documents. One such summariser is CENTRI- FUSER/PERSIVAL (Elhadad et al., 2005), which builds structured query-based representations of the documents as source for the summaries. The system was built using an iterative design that ac- commodates the feedback of a cohort of users. However, their developers acknowledge the lack of appropriate corpora, and to our knowledge nei- ther CENTRIFUSER nor PERSIVAL were tested on a specific corpus for comparison with other sys- tems. SemRep (Fiszman et al., 2004) provides ab- stractive summarisation by producing a semantic representation based on the UMLS concepts and their relations (Bodenreider, 2004) as found in the text. The evaluation was …

Content truncated to 500 words