Two reviewers individually and in duplicate reviewed the abstracts of 256 articles and selected 18 observational studies for inclusion.
Two reviewers (SDN, SA) independently and in duplicate reviewed the abstracts of 256 articles, and determined their eligibility for inclusion based on prespecified inclusion criteria as identified above.
The strengths of this review include the relatively broad scope of interventions examined, the use of systematic search strategies, duplicate reviews and duplicate abstraction of data, and the use of a Bayesian meta-analysis model designed for pooling of data from individually and cluster-randomized trials, and the examination of a clinically important outcome.
Two reviewers independently and in duplicate reviewed the titles and abstracts of retrieved publications and subsequently the full text of relevant articles.
The codebook was developed one domain at a time using a modified duplicate reviewing process that involved a set of primary reviewers extracting and documenting the information, and a second group of reviewers "auditing" (i.e., checking) primary reviews in small-group discussions; a third group of reviewers resolved disagreements.
They started by finding the near duplicate reviews, which they defined as reviews with a Jaccard similarity score of over 90%% of their 2-g.
A detailed approach in [18] identifies opinion spam as untruthful reviews, reviews on brand instead of product and advertisements and uses duplicate reviews as spam for classification purposes as well.
After independent review of all records in each group, the RNs discussed the duplicate reviewed records and reached consensus.
He then manually labeled the data by following a set of rules such as: duplicate and near duplicate reviews are labeled as spam, reviews about brands only are considered as spam, and non-reviews such as ads, discussions, or irrelevant reviews are labeled as spam.
A possibility for a less labor-intensive means of generating labeled training data is to find and label duplicate reviews as spam.
