Analyzing time series activity of Twitter political spambots

TL;DR

This study investigates whether temporal activity patterns can distinguish political Twitter bots from humans, finding that current bots mimic human behavior closely, thus challenging existing detection methods.

Contribution

The paper provides empirical evidence that current political Twitter bots closely imitate human activity patterns, highlighting the need for more advanced detection techniques.

Findings

No substantial differences in temporal patterns between bots and humans

Current political bots effectively mimic human behavior

Existing detection methods may be insufficient

Abstract

The presence and complexity of political Twitter bots has increased in recent years, making it a very difficult task to recognize these accounts from real, human users. We intended to provide an answer to the following question: are temporal patterns of activity qualitatively different in fake and human accounts? We collected a large sample of tweets during the post-electoral conflict in the US in 2020 and performed supervised and non-supervised statistical learning technique sto quantify the predictive power of time-series features for human-bot recognition. Our results show that there are no substantial differences, suggesting that political bots are nowadays very capable of mimicking human behaviour. This finding reveals the need for novel, more sophisticated bot-detection techniques.