Generating summary documents for a variable-quality PDF document collection

Abstract

The Cochrane Schizophrenia Group's Register of studies details all aspects of the effects of treating people with schizophrenia. It has been gathered over the last 20 years and consists of around 20,000 documents, overwhelmingly in PDF. Document collections of this sort -- on a given theme but gathered from a wide range of sources -- will generally have huge variability in the quality of the PDF, particularly with respect to the key property of text searchability. Summarising the results from the best of these papers, to allow evidence-based health care decision making, has so far been done by manually creating a summary document, starting from a visual inspection of the relevant PDF file. This labour-intensive process has resulted, to date, in only 4,000 of the papers being summarised -- with enormous duplication of effort and with many issues around the validity and reliability of the data extraction. This paper describes a pilot project to provide a computer-assisted framework in which any of the PDF documents could be searched for the occurrence of some 8,000 keywords and key phrases. Once keyword tagging has been completed the framework assists in the generation of a standard summary document, thereby greatly speeding up the production of these summaries. Early examples of the framework are described and its capabilities illustrated.