NLM at TREC 2012 Medical Records Track

摘要: Abstract : The NLM team used the relevance judgments for 2011 Medical Records track (that focused on finding patients eligible clinical studies) to analyze components of our systems. analysis showed that provided moderate improvements over baseline (established submitting topics as is Lucene) some and did not harm results any other topics. Our experiments confirmed implementing methods (such negation detection section splitting) motivated by text processing experience could improve identifying meet complex criteria inclusion in cohort studies. We therefore largely system with minor modifications document processing. submitted three automatic runs: an Essie run, two Lucene runs modifications. also interactive run which queries were interactively modified using until either top ten retrieved documents appeared mostly relevant or no be found. significantly outperformed all above medians submission types (achieving 0.37 infAP; 0.68 infNDCG; 0.75 P@10; 0.48 R-prec). Interestingly, values metrics common years this are very close achieved 2011. hypothetical overall-best best-manual performances better than run. topic frames web-based expansion (on but P@10 medians), it runs. medians. As 2011, we conclude existing search engines mature enough support selection tasks, quality