METHODS OF IMPLEMENTING DISINFORMATION DETECTION IN SOCIAL NETWORKS BASED ON ARTIFICIAL INTELLIGENCE

Abstract

The article considers modern approaches to automated detection of disinformation in social networks using artificial intelligence technologies. The evolution of methods is analyzed - from linguistic analysis of texts and classical machine learning algorithms to deep neural networks and transformative models. It is shown that traditional statistical methods do not provide the necessary accuracy when processing large amounts of data, while models based on CNN, RNN and BERT demonstrate high efficiency due to the ability to take into account context and semantic connections. Special attention is paid to multimodal analysis, which combines text, image and video processing to detect complex types of fakes, in particular deepfake. The use of the Leiden method is proposed as an innovative approach to clustering social graphs, which allows detecting coordinated communities of users spreading disinformation. An experimental study was conducted on data from the Twitter social network, which confirmed the high performance of the Leiden algorithm compared to the Louvain method. The obtained modularity (0.82) and cluster density (0.74) indicators demonstrated a clear structuring of disinformation communities and the ability to detect up to 78% of bot accounts. The developed model combines social graph analysis with natural language processing (NLP) methods to simultaneously identify sources of disinformation and the content of distributed messages. It is concluded that the integration of graph clustering methods and machine learning is a promising direction in creating automatic monitoring systems for social networks. Further research should focus on the development of explainable models (Explainable AI), multilingual adaptation, and the implementation of anti-fake technologies directly into the infrastructure of social platforms.