Information retrieval methodology for aiding scientific database search

During literature reviews, and specially when conducting systematic literature reviews, finding and screening relevant papers during scientific document search may involve managing and processing large amounts of unstructured text data. In those cases where the search topic is difficult to establish or has fuzzy limits, researchers require to broaden the scope of the search and, in consequence, data from retrieved scientific publications may become huge and uncorrelated. However, through a convenient analysis of these data the researcher may be able to discover new knowledge which may be hidden within the search output, thus exploring the limits of the search and enhancing the review scope. With that aim, this paper presents an iterative methodology that applies text mining and machine learning techniques to a downloaded corpus of abstracts from scientific databases, combining automatic processing algorithms with tools for supervised decision-making in an iterative process sustained on the researchers’ judgement, so as to adapt, screen and tune the search output. The paper ends showing a working example that employs a set of developed scripts that implement the different stages of the proposed methodology.