I’m not a human: A comparison of bot and human self-presentation

Social bots play a pivotal role in content creation and dissemination across social networks, often shouldering repetitive and mundane tasks. However, the emergence of malicious social bots has posed a significant threat to the online ecosystem, as they actively disseminate misinformation and disinformation. Many research efforts have been directed towards the detection of Twitter bots, a challenging endeavor given their non-identifiable nature through account names alone. This study takes a unique approach, focusing exclusively on the analysis of self-presentation by bots versus human users through linguistic analysis of written text on Twitter. We meticulously examine the linguistic characteristics of Twitter content generated by social bots and delve into the distinctions and commonalities within the content they produce. Employing a logistic regression model, we demonstrate the feasibility of distinguishing between text generated by bots and humans. Surprisingly, our findings reveal that bots generate content strikingly similar to that of humans, a phenomenon we elucidate through the lens of authorial agency theory. This research bears significant implications for social network designers, urging vigilance against the infiltration of malicious and fictitious accounts. Furthermore, it offers valuable insights into the characterization of social bots, which can be harnessed to enhance their detection mechanisms.