BOTTER: A framework to analyze social bots in Twitter

Abstract

Social networks have triumphed in communicating people online, but they have also been exploited to launch influence operations for manipulating society. The deployment of software-controlled accounts (e.g., social bots) has proven to be one of the most effective enablers for that purpose, and tools for their detection have been developed and widely adopted. However, the way to analyze these accounts and measure their impact is heterogeneous in the literature, where each case study performs unique measurements. To unify these efforts, we propose a common framework to analyze the interference of social bots in Twitter. The methodology compares the nonauthentic actors with the rest of users from different perspectives, thus building objective metrics to measure their actual impact. We validate the framework by applying it to a dataset of Twitter iterations dated in the weeks preceding the 2019 Spanish general election. In this sense, we check that our framework facilitates the quantitative evaluation of unauthentic groups, particularly discovering that social bots changed the natural dynamics of the network in these days, but did not have a significant impact. We also consider this methodology as a practical tool for the qualitative interpretation of experimental results, particularly suggesting within the aforementioned electoral context that semiautomated accounts are potentially more threatening than fully automated ones.

An Exploratory Study of Tweets about the SARS-CoV-2 Omicron Variant: Insights from Sentiment Analysis, Language Interpretation, Source Tracking, Type Classification, and Embedded URL Detection

“…Third, the in-built feature of Social Bearing that was used to remove bot-related content might not be the most optimal approach for removal of all bot-related tweets as certain advanced bot accounts that exactly mimic real user Twitter accounts in terms of Twitter user names and share/post content on Twitter after randomized time intervals (as opposed to the usual bots which usually share/post content after a fixed time interval) might be difficult to identify using this approach. Emerging works in the field of Natural Language Processing such as Tweez-Bot [151], Bot-DenseNet [152], Bot2Vec [153], Botter [154], and GlowWorm-based Generalized Regression [155] could be used for identifying such advanced bot accounts on Twitter. We could not implement any of these emerging works in our study due to the limited integration options provided by the Social Bearing research tool.…”

An Exploratory Study of Tweets about the SARS-CoV-2 Omicron Variant: Insights from Sentiment Analysis, Language Interpretation, Source Tracking, Type Classification, and Embedded URL Detection

“…Third, the in-built feature of Social Bearing that was used to remove potential bot-related content might not be the most optimal approach for the removal of all bot-related tweets, as certain advanced bot accounts that exactly mimic real user Twitter accounts in terms of Twitter usernames and share/post content on Twitter after randomized time intervals (as opposed to certain bots which share/post content after a fixed time interval) might be difficult to identify using this approach. Emerging works in the field of Natural Language Processing, such as TweezBot [151], Bot-DenseNet [152], Bot2Vec [153], Botter [154], and GlowWorm-based Generalized Regression [155], could be used for identifying such advanced bot accounts on Twitter. We could not implement any of these emerging works in our study due to the limited integration options provided by the Social Bearing research tool.…”