IEEE Technology and Society Magazine - September 2020 - 60

interests, location, education) and triad closure property to deceive them. In the infiltration process, socialbots target attributes or structurally coherent users,
raising the probability of acceptance of the friendship requests.

Detection Challenges
To tackle the socialbot problem, researchers from academia and industry have proposed various approaches
and are still working on this problem. We group the
existing approaches into five categories, namely
machine learning, graph-based, behavior- based, crowdsourcing-based, and hybrid approaches.
In machine learning approaches, classification models using a set of predefined features are trained to predict the label of a new user [22]. As new features are
devised and classification models are updated, socialbots change their behavior to bypass the detection systems, making detection a challenging problem. Machine
learning approaches use handcrafted features, which is
a manual and time-consuming task. As a result, it incorporates human biases and deficiencies.
Moreover, the efficacy of a machine learning-based
classification system depends on a set of defined features. Therefore, if those defined features are of low
quality, the performance of the resulting trained classification system automatically will be low. Graph-based
approaches partition a social graph into multiple subgraphs representing sybil and non-sybil regions [29].
Socialbots deceive this line of techniques by creating
attack edges with benign users. Further, graph-based
approaches do not guarantee the detection of an individual socialbot.
Researchers have also modeled the operational and
behavioral patterns of users to differentiate between
malicious and benign users [21], [30]. These approaches model synchronization among user activities to
detect suspicious behavior. Presently, these approaches
appear most effective in comparison to other categories
of approaches. In a crowd-sourcing approach, expert
annotators are hired to label an account as a socialbot
or benign, based on its profile and activity information.
However, this is neither economically nor technically
feasible for OSNs, as they have millions of users [5].
Researchers have also proposed hybrid approaches for
socialbots detection.

Current Direction of Socialbot Research
and its Role in COVID-19
The focal point of earlier socialbot research was on
impact and infiltration analysis [7], [24], [25], and development of detection systems [21], [22], [30]. Recently,
researchers have started exploring the role of socialbots
in the context of various threat dimensions, such as

political tools [31], [32], fake news and rumor spreaders
[4], [33], and trolls [12], [34], as depicted in Figure 6.
Recently, bots have also been also misused in diffusing public health misinformation (i.e, infodemics) and
amplifying vaccination debates [35]-[37]. In the global
coronavirus (COVID-19) crisis, most of the countries
around the world are under lockdown to contain the
spread of the virus. As a result, people are staying at
home and spending more time on social media platforms. Researchers are conducting studies to observe
insights from user-generated OSN content in the context
of COVID-19. In such a study, Ferrara [38] analyzed the
role of socialbots and found their use for both good and
malicious purposes. The author observed the use of
socialbots to foster democratic discussions, which may
otherwise be censored. On the other hand, the author
also found the use of socialbots in promoting divisive
political campaigns and conspiracy theories. However, a
comprehensive analysis regarding the role of bots in diffusing different types of weaponized information like
artificial political campaigns, health infodemics, xenophobic behavior, rumor, and fake news in the context of
COVID-19 is not yet adequately studied. These are good
directions for research to investigate the contextual role
of the socialbots.

Socialbot Problem Still Evolving
Based on the discussions and analysis presented in this
article, we can conclude that the socialbot problem is
still evolving and targeting new areas. Existing literature
is also lacking in terms of efficient techniques for prevention and detection of the sophisticated version of
socialbots like political bots, influence bots, and spambots. In this paper, we have presented a study of socialbots from four different perspectives - their
differentiating characteristics with Web bots, their infiltration ability on Twitter, their different threat-dimensions, and various categories of defense challenges. We
have also presented a comprehensive discussion
regarding the role of socialbots in the contexts of various OSN threats. Finally, we have presented a brief overview of the current trends in socialbots research and
their role in the context of the COVID-19 pandemic. We
aim for this article to be enlightening and informative
for a comprehensive understanding of socialbots and
related research challenges.

Author Information
Muhammad Abulaish is with South Asian University,
Department of Computer Science, Akbar Bhawan
Chanakyapuri, New Delhi, Delhi 110021, India. Email:
abulaish@sau.ac.in.
Mohd Fazil is with Jamia Millia Islamia, Department of
Computer Science, New Delhi, Delhi, India.

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

∕

SEPTEMBER 2020

IEEE Technology and Society Magazine - September 2020

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2020