Uncovering Social Media Bots: a Transparency-focused Approach

As the Online Social Networks (OSNs) presence continues to grow as a form of mass communication, tensions regarding their usage and perception by different social groups are reaching a turning point. The number of messages that are exchanged between users in these environments are vast and brought a trust problem, where it is difficult to know if the information is from a real person and if what was said is true. Automated users (bots) are part of this issue, as they may be used to spread false and/or harmful messages through an OSN while pretending to be a person. New attempts to automatically identify bots are in constant development, but so are the mechanisms to elude detection. We believe that teaching the user to identify a bot message is an important step in maintaining the credibility of content on social media. In this study, we developed an analysis tool, based on media literacy considerations, that helps the ordinary user to recognize a bot message using only textual features. Instead of simply classifying a user as a bot or human, this tool presents an interpretable reasoning path that helps to educate the user into recognizing suspicious activity. Experimental evaluation is conducted to test the tool’s primary effectiveness (classification) and results are presented. The secondary effectiveness (interpretability) is discussed in qualitative terms.