Determining markers of author identification on short texts: A forensic linguistic application on Twitter

According to the literature, identification and/or profiling of the author of a given text is one of the most popular questions addressed to forensic linguists. For this very reason, it could be stated that author identification studies are of particular importance in terms of forensic linguistics. Besides, such studies may either be performed individually or have a supplementary essence along with other several analyses. Despite all, author identification studies concerning Turkish texts have only recently received scientific attention and within this context, short texts have usually been disregarded. It is clear that there is a need for specific methods accounting for texts consisting limited linguistic data, with a special regard to correspondence on mobile phones and on social media posts. Most of the features that are employed in the author identification analyses of newspaper articles seem to be inadequate on social media posts or other instant messaging types since the latter sort of texts have considerably limited verbal content. This study aims to address to this necessity within the given context. First, a total of 1000 posts by 50 different authors from the micro-blog site Twitter have been examined as a sample of short texts and then, distinctive features have been established which are useful to distinguish the same and different authors. These features that are sorted out under titles "spelling and punctuation", "lexis and context" and "others" have been worked out on a total of 198 single Twitter posts submitted by 60 different authors. Through 18 separate analyses that include different number of authors and posts, identification of the initially unknown authors have been attempted. Every single attempt resulted with a match and these matchups were all checked and confirmed. There hasn't been any analysis resulted with indecision or misdetection. Findings seem to be significant for both constituting a methodology for author identification studies on short texts in Turkish, as well as for providing a basis for the advancement of the proposed method in the future.