Short video platforms have rapidly emerged as a primary source of news consumption, offering engaging and easily digestible content. However, their widespread popularity has also facilitated the proliferation of fake news, posing a significant societal concern and exacerbating the challenge of misinformation control. Unlike traditional text-based news, fake news on short video platforms often spreads through a combination of textual, visual, and auditory elements, making detection particularly complex. Existing methods leveraging multimodal features struggle to capture deep semantic associations and lack effective fusion mechanisms, resulting in limited cross-modal interactions and suboptimal feature representation. Furthermore, simple feature concatenation or linear weighting may fail to fully exploit the complementary information across modalities. To address these challenges, we propose DAFSVFND, a novel multimodal fusion network that incorporates a stacked text-guided dual attention mechanism to facilitate deep interaction and fusion among textual, visual, and auditory features. By strengthening cross-modal dependencies and refining feature representations, our approach enhances the model’s ability to distinguish between real and fake news. Extensive experiments on two large-scale fake news datasets demonstrate that our model significantly outperforms state-of-the-art methods, offering an effective solution for fake news detection on short video platforms.