Speaker
Description
Bello, S.
DEPARTMENT OF COMPUTER SCIENCE EDUCATION ISA KAITA COLLEGE OF EDUCATION DUTSIN-MA, KATSINA STATE, NIGERIA
kurami01@gmail.com
Track: Research, Innovation and Knowledge Ecosystems
ABSTRACT 5
The rapid proliferation of AI-generated deepfake videos on short-form platforms such as TikTok poses an emerging threat to digital integrity, particularly within higher education contexts where learners increasingly consume user-generated content as informal learning resources. Existing deepfake detection methods remain largely unimodal, brittle under real-world compression, and disconnected from fact-checking mechanisms leaving educational ecosystems vulnerable to persuasive misinformation. This study addresses these gaps by designing, developing, and empirically validating a novel Multimodal and Robust Deepfake Detection Framework (MRDF) tailored for short-form video environments. Grounded in Design Science Research, the proposed framework integrates three core modalities visual (facial and motion artifacts), audio (speech synthesis anomalies), and textual (captions and transcribed claims) using attention-based fusion to capture cross-modal inconsistencies. Robustness is achieved through platform-aware simulation of compression, noise, and adversarial perturbations, while a retrieval-augmented semantic verification module assesses factual consistency against trusted knowledge bases (e.g., WHO, Africa Check). Evaluation using benchmark datasets (FaceForensics++, DFDC, FakeAVCeleb) and platform-simulated conditions demonstrates that MRDF significantly outperforms unimodal and non-robust baselines in detection accuracy, cross-dataset generalization, and resilience to distortion. Furthermore, the framework’s explainability module provides interpretable outputs, supporting human-in-the-loop moderation. For higher education institutions, MRDF offers a blueprint for next-generation content authentication systems, enabling educators, librarians, and administrators to identify AI-generated misinformation before it reaches learners. By bridging technical detection with semantic verification, this research contributes to building future-ready, trustworthy digital learning ecosystems across African universities and beyond.
Keywords: Deepfake detection, multimodal learning, short-form video platforms, AI misinformation, higher education, TikTok, fact-checking, robust AI