| dc.description.abstract | Currently we can find numerous tools to extract metadata from scientific papers, each one with its own particularity, technology and used techniques. However, with the increasing scientific production and the numerous publishers, events and conferences, a large part of papers still remain without an effective automated metadata extraction, hindering theknowledge dissemination and mainly the electronic search for these documents. The present work compares the correct metadata extraction from some preselected tools - Cermine, CiteSeer, CrossRef and ParsCit - using an empirical experiment with a set of scientific papers. This set covers different knowledge areas, conferences and different layouts. The experiment was made inside custom environments according the technologies each tool needs, allowing all papers to have their metadata extracted by each one and comparing results one by one. Thereby, according the presented results, its possible to identify the behavior of each tool related to the right metadata extraction. Except for the CrossRef tool, all others obtained results over 60%, including Cermine with to 86.83%. Moreover, the major weaknesses were identified for each tool; points to be fixed; metadata fields with better extraction results. Furthermore its also presented a reliability index, that allows to establish a grade for each tool. This grade is calculated based on the metadata extraction results using the selected set of papers. | |
