Twitter features distributions across similar labelers

This investigation was one part of a study that involved modelling the credibility of Arabic microblogs with disagreed judging credibility labels. We investigated the hypothesis that the most similar and agreed microblogs credibility judges use the same Twitter credibility features to evaluate tweet messages. First, the most similar labelers were identified using different similarity and agreement measures. Then, we used their assigned credibility labels to examine Twitter content and author features distributions. We found that similar and agreed labelers did not continuously correlate with assigning credibility judgments labels harnessing the same Twitter features. Shared similar features between labelers mainly appeared with the most agreed labelers using Krippendorffs alpha coefficient measure resulting in close to 100% features similarity for only the low-credibility class.