https://jjcit.org/paper/230 10.5455/jjcit.71-1709201313 Nataliya Boyko Technologies,Information environment,Neural networks,Testing approaches,Disinformation sources 2 2780 367 29-Feb.-2024 26-Apr.-2024 14-May-2024 In the context of the growing popularity of social media over the past ten years, an urgent problem of fake news spreading has arisen, which underscores the research’s relevance. The aim of this article is to assess the efficacy of multimodal approaches in detecting fake news, a pressing issue given the substantial impact misinformation can have on health, politics and economics. To achieve this goal, a multimodal approach was chosen that combines deep-learning frameworks and pre-trained models. This approach provides a comprehensive analysis of textual, visual and audio information, allowing for more accurate identification of disinformation sources. The use of various knowledge-transfer methods made it possible to process information efficiently, improving the quality of classification. The study conducted a thorough analysis of various data-collection strategies, as well as a comparative analysis of available multimodal approaches to fake-news detection and the datasets used. The results of this study included a detailed analysis of current research work in the field of fake-news detection and the development of a multimodal approach to this problem. Textual, visual and audio information was processed using pre-trained models and deep learning, achieving high accuracy in fake news detection. The results of the study indicated that the multimodal approach allows for more accurate identification of sources of disinformation and increases the efficiency of fake-news classification compared to other methods. A comparative analysis of various data collection strategies and datasets was also conducted, confirming the high efficiency of the approach under various conditions.