Detection Technology of Social Robot: Based on the Interpretation of Botometer Model

摘要

In the era of Web 2.0, social media have been a significant place for democratic conversation about social or political issues. While in many major public events like the Russia-Ukraine war or U.S. Presidential election, enormous social bots were found on Twitter and Facebook, putting forward public opinion warfare. By creating the illusion of grassroots support for a certain opinion, this kind of artificial intelligence can be exploited to spread misinformation, change the public perception of political entities or even promote terrorist propaganda. As a result of that, exploiting detection tools has been a great concern since social bots were born. In this article, we focused on Botometer, a publicly available detection tool, to further explain the AI technologies used in identifying artificial accounts. By analyzing its database and combing the previous literature, we explained the model from the aspect of data augmentation, feature engineering, account characterization, and Ensemble of Specialized Classifier (ESC). Considering the consistent evolution of social bots, we propose several optimization suggestions and three other techniques or models to improve the accuracy of social bots detection.