Google versus other text similarity tools in detection of plagiarism: a pilot study in the Journal of Clinical and Diagnostic Research

Abstract

A study conducted by Baždarić and colleagues 4 in 2012 showed that 14% of manuscripts submitted to the Croatian Medical Journal were suspected of plagiarism. However, on manual verification, 11% of these submissions were considered plagiarised. Therefore, to avoid false positive results, a manual verification by an editor is required.We practised using CrossCheck (iThenticate) in JCDR, to check for plagiarism, but after a few instances where we could identify misses, we re-evaluated our strategy. Later, we started to manually search for similar sentences using Google. We conceptualised a study to bring out any significant gain that we might have had by this change in our practice against the increased manual work. For the completion of the study we considered eight software programmes, 2 ran a feasibility test and found that three of them were easily assessable and usable. Our primary aim was to compare three text similarity programmes (iThenticate, Plagiarism Checker X, and Viper) against Google in detecting text similarity in our journal. MethodsSince 2014, we have employed an in-house staff member to check for plagiarism using Google. The staff member was asked to provide all manuscripts, submitted from 1st January to 31st May, 2015, which were flagged during the initial screenings with markedly high text similarity.JCDR received 1700 manuscripts in this time period. In 96 manuscripts, the decision was greatly affected due to the presence of text similarity. From these, 25 manuscripts (16 original manuscripts and 9 case reports) were randomly selected. We do not usually quantify the text similarity into percentages. The chunks of similar text are marked and the URL links of the source are tagged along the manuscript. In certain cases, the editor rejects a manuscript exclusively based on this so-called Google report and a review report where the reviewer advises rejection of the manuscript. In both situations, the text similarity is considered as an alarm for the decision on the manuscript-either major revision or rejection. The editorial policy is to remove the text similarities by rephrasing the sentences and citing the references, if the original article or case-report quality of data outweighs the draft originality. AbstractBackground: We practised using plagiarism detection software in the Journal of Clinical and Diagnostic Research, but after a few significant items were missed, we re-assessed our strategy and compared Google with three other text similarity programmes.Method: 25 manuscripts (16 original articles and 9 case reports) were randomly selected, where the decision to publish had been greatly affected by plagiarism. These manuscripts were checked for plagiarism, searching each sentence using Google. The same manuscripts were run through three text similarity software programmes (iThenticate, Viper, and Plagiarism Checker X). For original research, we considered methodology, results, and discussion; and for case reports, we considered case details and discussion. Each report was checked by the i...