Skip to main content

DOCTORAL DEFENSE: Automating Social Interactions: A network perspective of the role of bots on social media

CEU Vienna
Wednesday, December 6, 2023, 3:00 pm – 5:00 pm

Please note that registration for the event closes at 2:00pm CET on December 6. The meeting link will be sent to all registered attendees shortly after 2:00pm.

Members of the Dissertation Committee:

Chair: Petra Kralj Novak (CEU DNDS, voting)

Supervisor: János Kertész (CEU DNDS, non-voting)

Reviewer: Alaistair Reed (Swansea University, voting)

Reviewer: Onur Varol (Sabanci University, voting)

ABSTRACT / This thesis comprises three projects examining the emerged network structures due to bot activities on Twitter and Telegram. Specifically, we delve into these structures' origins, impacts, and roles in a broader network context. Our analysis uses data-driven, experimental, and network-based concepts to identify varying network structures across contexts and user groups, such as core-periphery structures, local structures (i.e., motifs), community formations, and rich-club organization.

The first project is an experimental study on Twitter where we deployed six bots in pairs of two, each pair assigned different strategies: a trend-targeting strategy (TTS), a keywords-targeting strategy (KTS), and a user-targeting strategy (UTS). We then assessed interaction patterns, including targeting users, message dissemination, relationship propagation, and engagement. While TTS was the most effective in obtaining human feedback, it displayed the least diverse local structure patterns. In contrast, UTS was the least effective but activated a broader spectrum of complex, local structures. Furthermore, content-related strategies (TTS and KTS) had a significant overlap in terms of local structures activated. Notably, the KTS shows promise in bridging the benefits of content-focused and user-focused approaches by targeting content that resonates with particular users. This strategy has shown the ability to create engaging patterns while effectively disseminating content, which is vital to success on social media platforms.

The second project is concerned with the network structures of three extremist groups on Telegram: the Islamic State of Iraq and Syria (ISIS), far-right groups (FR), and pro-Russian actors tied to the conflict in Ukraine (PR). We expected different authority structures: ISIS lacked a centralized authority, the pro-Russian actors displayed a pronounced central authority, and the far-right group combined decentralized and centralized elements. Network metrics-based analysis supported the expectations that the three extremist groups follow different organizational principles and platform usage purposes, which results in different structures. Our application of the 'rich club' detection method disclosed variations in the nodes' roles and positions within these networks. Bots are present in the rich-club of the ISIS and FR networks, while the PR network's rich-club is exclusively composed of human users. Given their automatic nature, bots could increase the pace of information spread within the network, but in a very centralized network, there is no need for such an augmentation.

The third project identified two primary communities of bots and channels/groups associated with ISIS on Telegram. These basic bots, notwithstanding their simplicity, remain pivotal in sustaining the online presence of the Islamic State, especially in the light of Telegram's intensive countermeasures. Furthermore, the core of both communities mainly consists of bots, with their peripheries comprising a mixture of channels and groups. A functional explanation is that the core-periphery structures have emerged because of constant activity from the core in maintaining content distribution efforts and chat moderation, which was conducted in the periphery.

BIO / Abdullah is a doctoral candidate specializing in Network Science at the Central European University, expanding upon his academic foundations after obtaining a master’s degree in engineering. His primary research focus lies in the practical application of data science, with a particular emphasis on investigating the propagation of disinformation and extremist content across Online Social Networks (OSNs). His work also extends to the network modeling of interactions between humans and bots within these social platforms. Abdullah has developed a keen interest in understanding the role and impact of Large Language Models (LLMs) in social scenarios. Currently, his research is focusing on the dynamics between social and political bots and human users within instant messaging applications such as Telegram. In addition to this, Abdullah is exploring the nature and implications of LLMs, while also studying the structures and behaviors of extremist networks.