University of Twente Student Theses
On the Evolution of Social Bot Intelligence
Bouman, S.M. (2025) On the Evolution of Social Bot Intelligence.
|
PDF
11MB |
| Abstract: | In 2020, Assenmacher et al. investigated the theoretical and practically achieved sophistication of social bots by analysing metadata from code repositories on public code sharing platforms in their paper “Demystifying Social Bots; On the Intelligence of Automated Social Media Actors". They found a significant gap between what intelligent technologies social bots could use in theory and the predominantly simple technologies used in practice. Five years later, after the huge boom in both popularity and accessibility of Large Language Model (LLM) technologies, this has changed. The aim of this work was to map out the current landscape of social media bots on GitHub and procure empirical evidence of how, if at all, LLM technologies are being used by social bots. This thesis shows the developments in the overall ecosystem of social media bots on GitHub, finding that largely thanks to bots for WhatsApp and Telegram, the number of repositories has tripled and is increasingly growing. Manual analysis of popular repositories as well as advanced topic modelling of all repository descriptions exposes a large variety of social media bots on GitHub, both in terms of purpose and in terms of technologies used to create them. Furthermore, it is explored how publicly available social bot repositories can be used to gain insight into the practical use of LLM technologies by social bots. By exploiting multiple novel analysis methods, including the use of Google Gemini Pro to analyse source code, this thesis finds that an estimated 10.58% of social bot repositories use LLM technologies across nine different social media platforms. The majority of these (almost all benign) social bots utilise accessible LLM Application Program Interfaces (APIs), like the one maintained by OpenAI. The bots use LLM-generated content in both one-to-one and one-to-many communications, are generally capable of tracking conversation history and using custom system prompts, and in some cases can obfuscate their identity as a bot. |
| Item Type: | Student Thesis (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science, 70 social sciences in general |
| Programme: | Computer Science MSc (60300) |
| Link to this item: | https://purl.utwente.nl/essays/106664 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page