Paper Type
Complete
Paper Number
PACIS2026-1146
Description
This paper discusses deepfakes in social media, a phenomenon that poses a critical challenge to digital trust and information integrity. The deepfakes are improving as technical artifacts, but a cohesive synthesis of the literature is lacking. To address this gap, we conducted a thematic analysis of 109 relevant articles from the literature (2018–2026) that examine the socio-technical implications of deepfakes. Our methodology introduces a rigorous cross-model synthesis through Large Language Models (LLMs). The emerged themes are subsequently mapped onto a corpus of online news articles extracted via GDelt API, using GLDA topic modeling to analogize scholarly findings with real-world insights. The study further identifies critical research gaps in the deepfake domain as it overcommits reactive (forensic detection) and neglects anticipatory (digital literacy) approaches, leading to a Structural Reactivity Gap. This provides the theoretical basis for shifting the deepfake research paradigm from artifact-centric (detecting deepfakes) to agent-centric (understanding deepfakes).
Recommended Citation
KUMAR, PARVEEN and Dixit, Gaurav, "LLMs-based Thematic Analysis of Social Media Deepfakes and its Empirical Parallelism with Deepfake News" (2026). PACIS 2026 Proceedings. 2.
https://aisel.aisnet.org/pacis2026/ai_ml/ai_ml/2
LLMs-based Thematic Analysis of Social Media Deepfakes and its Empirical Parallelism with Deepfake News
This paper discusses deepfakes in social media, a phenomenon that poses a critical challenge to digital trust and information integrity. The deepfakes are improving as technical artifacts, but a cohesive synthesis of the literature is lacking. To address this gap, we conducted a thematic analysis of 109 relevant articles from the literature (2018–2026) that examine the socio-technical implications of deepfakes. Our methodology introduces a rigorous cross-model synthesis through Large Language Models (LLMs). The emerged themes are subsequently mapped onto a corpus of online news articles extracted via GDelt API, using GLDA topic modeling to analogize scholarly findings with real-world insights. The study further identifies critical research gaps in the deepfake domain as it overcommits reactive (forensic detection) and neglects anticipatory (digital literacy) approaches, leading to a Structural Reactivity Gap. This provides the theoretical basis for shifting the deepfake research paradigm from artifact-centric (detecting deepfakes) to agent-centric (understanding deepfakes).
Comments
01-AIML