Its all in a name: detecting and labeling bots by their name

Abstract

Automated social media bots have existed almost as long as the social media environments they inhabit. Their emergence has triggered numerous research efforts to develop increasingly sophisticated means to detect these accounts. These efforts have resulted in a cat and mouse cycle in which detection algorithms evolve trying to keep up with ever evolving bots. As part of this continued evolution, our research proposes a multi-model 'tool-box' approach in order to conduct detection at various tiers of data granularity. To support this toolbox approach this research also uses random string detection applied to user names to filter twitter streams for bot accounts and use this as labeled training data for follow on research.

Detecting Political Bots on Twitter during the 2019 Finnish Parliamentary Election

“…Both supervised [6,15] and unsupervised [16,21] machine learning models have been used in bot detection research. The drawback of supervised learning is that creating a labeled dataset for training the model either requires a large amount of manual labeling [6] or using a pre-labeled dataset, which may limit the applicability of the model as the datasets most likely represent only a fraction of the possible behavior of bot accounts in Twitter.…”

“…Past studies on bot detection have been to some extent restricted to bots with a specific feature. For instance, Beskow and Carley [15] managed to identify specific automatically generated bot accounts based on a single piece of metadata, the profile name, with approximately 95%-99% accuracy depending on the algorithm used. However, this type of approach results in a very narrow use and the aforementioned model could only detect bot accounts that have an account name consisting of a randomly generated string of 15 characters and more than likely to miss out the bots with different characteristics.…”

“…However, this type of approach results in a very narrow use and the aforementioned model could only detect bot accounts that have an account name consisting of a randomly generated string of 15 characters and more than likely to miss out the bots with different characteristics. However, as Beskow and Carley [15] propose, a tool-box approach where multiple different models are combined can make even the simple models an important contribution to more advanced bot detection models.…”

“…Profiles that have none or a few followers, but follow many profiles are suspect [8]. Lastly, the contents of the textual metadata can be analyzed and used to classify bots for instance by inspecting the length or frequency of certain keywords in the description or name [8,15,18].…”

“…Features in this group are occasionally used in earlier studies. Two of the features in this list simply check whether an account is a bot according to the profile description or name by looking if the fields contain the word "bot" [15]. The rest of the features relate to an examination of URLs, hashtags ,or other keywords [20].…”