Extended Boolean retrieval for systematic biomedical reviews

Searching for relevant documents is a laborious task involved in preparing systematic reviews of biomedical literature. Currently, complex Boolean queries are iteratively developed, and then each document of the final query result is assessed for relevance. However, the result set sizes of these queries are hard to control, and in practice it is difficult to balance the competing desires to keep result sets to a manageable volume, and yet not exclude relevant documents from consideration. Ranking overcomes these problems by allowing the user to choose the number of documents to be inspected. However, previous work did not show significant improvements over the Boolean approach when ranked keyword queries based on terms in the Boolean queries, review title, research question or inclusion criteria were used. The extended Boolean retrieval model also provides ranked output, but existing complex Boolean queries can be directly used as formal description of the complex information needs occurring in this domain. In this paper we show that extended Boolean retrieval is able to find a larger quantity of relevant documents than previous approaches when comparable (or greater) numbers of documents are inspected for relevance.