• Anshul

The rapid growth of online video platforms has increased the spread of fabricated and manipulated news content. Unlike traditional text-based misinformation, fake news videos can contain misleading information across multiple modalities, including visual frames and spoken content. This paper presents TruthLens, a lightweight multimodal framework for detecting fake news videos by combining visual and textual information. The proposed system uses ResNet50 to extract visual features from sampled video frames, Whisper for automatic speech transcription, BERT for contextual textual feature extraction, and XGBoost for final classification. The extracted 2048-dimensional visual representation and 768dimensional textual representation are concatenated to produce a 2816-dimensional multimodal feature vector. The current dataset contains 305 videos, consisting of 180 REAL and 125 FAKE samples. Each video is processed using eight sampled frames, audio transcription, and multimodal feature extraction. On the held-out test set, the system achieved 82.61% accuracy, 82.35% precision, 73.68% recall, 77.78% F1score, and 0.961 ROC-AUC using the reference threshold of 0.50. A threshold of 0.45 was previously selected during validation experiments, while the current production and reporting decision threshold is 0.30 (30%). To improve interpretability, SHAP is used for feature-level explanation and Grad-CAM is used to visualize influential regions within video frames. The proposed framework is intended to evaluate generalization to unseen fake video types and sources rather than claim universal detection of all possible fake videos.

Fake News Detection, Fake News Videos, Multimodal Learning, Video Analysis, ResNet50, Whisper, BERT, XGBoost, Explainable Artificial Intelligence, SHAP, Grad-CAM, Social Media Misinformation.

"TruthLens: A Lightweight Multimodal Framework for Fake News Video Detec", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 9, page no.c388-c397, September-2026, Available :http://www.jetir.org/papers/JETIR2609243.pdf

"TruthLens: A Lightweight Multimodal Framework for Fake News Video Detec", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 9, page no. ppc388-c397, September-2026, Available at : http://www.jetir.org/papers/JETIR2609243.pdf