Mining for Relevant Literature from the MEDLINE Database Using Keyword Scoring

MOTIVATION: It would be useful to be able to retrieve a ranked publication list relevant to topics of interest. METHODOLOGY: This novel approach uses semantics built into the Medical SubHeading (MeSH) hierarchy. A training set of papers is constructed, each annotated with MeSH terms of interest (e.g. Stem Cell or some child in its subtree). Frequencies of nouns are computed, and ratios calculated of noun frequencies in the training set to frequencies in MEDLINE. Ratios scored abstracts in the training set, and on an equal number of papers randomly selected from MEDLINE. Performance of the relevance-prediction algorithm was benchmarked by counting Stem Cell papers retrieved after merging ranked lists of scored papers and selecting increasing numbers of top papers. RESULTS: 16 MeSH keywords were used to build a training set of 99,020 papers known to be relevant to Stem Cells. 43,328 and 11,773 nouns were extracted from MEDLINE and the training set, respectively. From the training set, 2,078 keywords occurred more than 100 times in the literature, and top-scoring keywords were verified to be relevant to stem cells. Benchmarking the relevance-prediction algorithm reported an 88% recall and precision of stem cell papers when half of the merged list was selected. 3 Acknowledgments I would like to thank my advisors Dr. Miguel Andrade and Dr. Tony White for their guidance and advice; my colleagues Dr. for their patience with my infinite stream of newbie questions; my parents for bringing me into this world and always encouraging me towards becoming an outstanding member of society; and last but not least Ms. Evelyne Chevrier, for inspiring me when I need it most. 4 Table of Contents 1 Introduction 6 1.1 Goals of Literature Mining 6 1.2 Progress in the field of Literature Mining 7 1.3 Use of Keywords and Hierarchies 13 1. Methodology 16 2.1 Gather databases: 16 2.2 Generate a list of stem cell MeSH keywords: 16 2.3 Extract papers about stem cells: 16 2.4 XML document validation: 17 2.5 Moving a range of files: 17 2.6 Robustly extract papers about stem cells: 17 2.7 Combine lists of stem cell papers: 17 2.8 Extract nouns from MEDLINE papers. 18 2.9 Extract nouns from stem cell papers: 18 2.10 Compute the contrast between stem cell papers and the rest of MEDLINE: 18 2.11 Score papers: 18 2.12 Benchmarking: 19 3 Results 20 3.2 Generate a list of stem cell MeSH keywords: 20 3.7 …