Seeking Informativeness in Literature Based Discovery

Abstract

The continuously increasing number of publications within the biomedical domain has fuelled the creation of literature based discovery (LBD) systems which identify unconnected pieces of knowledge appearing in separate literatures which can be combined to make new discoveries. Without filtering, the amount of hidden knowledge found is vast due to noise, making it impractical for a researcher to examine, or clinically evaluate, the potential discoveries. We present a number of filtering techniques, including two which exploit the LBD system itself rather than being based on a statistical or manual examination of document collections, and we demonstrate usefulness via replication of known discoveries.