Relevance Feedback Query Refinement for PDF Medical Journal Articles

Abstract

This paper addresses relevance feedback as an alternative to keyword-based search engines for sifting through large PDF document collections and extracting the most relevant documents (especially for literature review purposes). Until now, relevance feedback has only been used in content-based image and video retrieval due to the inability to query those media types without keywords. Since PDF journal articles contain many valuable non-keyword features such as structure and formatting information as well as embedded figures, they would benefit from relevance feedback. Stripping a PDF into "full-text" for indexing purposes disregards these important features. We discuss how they can be used to our advantage and look to integrate the wealth of knowledge from relevance feedback text-based information retrieval. We argue for the benefits of placing the burden of relevance judgement on the user rather than the retrieval system and present alternative document views that quickly allow the user to deem relevance.