Abstract
Translational neuroscience in the field of spinal cord injuries (SCI) faces a strong disproportion between immense preclinical research efforts and a lack of therapeutic approaches successful in human patients: Currently, preclinical research on SCI yields more than 3,000 new publications per year (8,000 when including the whole central nervous system, growing at an exponential rate), whereas none of the resulting therapeutic concepts has led to functional recovery of neural tissue in humans. Improving clinical researchers’ information access therefore carries the potential to support more effective selection of promising therapy candidates from preclinical studies. Thus, automated information extraction from scientific publications contributes to enabling meta studies and therapy grading by aggregating relevant information from the entire body of previous work on SCI.
We present SCIE, an automated information extraction pipeline capable of detecting relevant information in SCI publications based on ontological entity and probabilistic relation detection. The input are plain text or PDF documents. As output, the user choses between an online visualization or a machine-readable format. Compared to human gold standard annotations, our system achieves an average extraction performance of 76 % precision and 52 % recall (F1-measure 0.59).
An instance of the webservice is available at http://scie.sc.cit-ec.uni-bielefeld.de/. SCIE is free software licensed under the AGPL and can be downloaded for local installation at http://opensource.cit-ec.de/projects/scie/.
