DEVELOPMENT OF METHODS FOR DETECTION OF FAKE NEWS BASED ON GRAPH THEORY ANALYSIS

Abstract

The article is devoted to the development and research of methods for automated detection of fake news based on the analysis of graph structures and algorithms for detecting communities in social networks. The relevance of the study is due to the rapid growth of the volume of news content and the complexity of operational verification of its reliability in the digital information environment. It is shown that traditional approaches based exclusively on linguistic analysis of texts and classical machine learning methods are insufficient for detecting coordinated disinformation campaigns that have a network nature of distribution. The paper substantiates the feasibility of using a graph model of presenting news data, in which individual messages and information entities are considered as vertices of a graph, and their relationships as edges. This approach allows us to move from the analysis of isolated news to the study of structural patterns of information flows. The main focus is on the application of Louvain and Leiden community detection algorithms, which provide graph clustering based on modularity optimization and allow identifying densely connected groups of messages. The architecture of a software tool for detecting fake news is proposed, which includes the stages of data loading and preprocessing, graph model construction, clustering, analysis of results, and visualization. The implementation is carried out according to the modular principle, which provides flexibility, scalability, and the ability to integrate alternative clustering algorithms. As part of the experimental study, a comparative analysis of the results obtained using the Louvain and Leiden algorithms was conducted on real social network data. The experimental results showed that both algorithms are capable of forming communities with a high level of purity, but the Leiden algorithm provides more stable and internally connected clusters, which increases the interpretability of the results and the accuracy of identifying disinformation groups. It is demonstrated that the combination of text similarity analysis with graph communities allows for more effective detection of fake news and coordinated information influences compared to approaches focused only on content. The practical value of the work lies in the creation of a software tool with a graphical user interface that provides a transparent and reproducible process of analyzing news messages in real time. The proposed approach can be used as a basis for disinformation monitoring systems, decision support in the field of information security and further scientific research. Prospects for further work are related to the use of dynamic graphs, multimodal data and explainable artificial intelligence models for in-depth analysis of fake news dissemination processes.