Devising an appropriate Search String for a secondary study is not a trivial task and identifying suitable keywords has been reported in the literature as a difficulty faced by researchers. A poorly chosen Search String may compromise the quality of the secondary study, by missing relevant studies or leading to overwork in subsequent steps of the secondary study, in case irrelevant studies are selected. In this paper, we propose an approach for the creation and calibration of a Search String. We chose three published systematic literature reviews (SLRs) from Scopus and applied Machine Learning algorithms to create the corresponding Search Strings to be used in the SLRs. Comparison of results obtained with those published in previous SLRs, show an increase of recall of revisions by up to 12%, with no loss of recall. To motivate future studies and replications, the tool implementing the proposed approach is available in a public repository, along with the dataset used in this paper.

--

Secondary studies,Machine learning,Natural language processing

• Ciências da Computação e da Informação - Ciências Naturais