DETECTION OF THE MOST COMMON FAKE NEWS USING LEIDEN CLUSTERING

Abstract

Today, a significant challenge is the automatic identification of fake news and the detection of groups of messages that spread unreliable or manipulative content. One of the effective tools for addressing this task is the Leiden clustering method. The Leiden algorithm belongs to community detection methods in graphs and enables grouping news items based on their internal similarity. For news analysis, a set of characteristics is first formed for each message: its textual content, semantic features, source of dissemination, publication time, and other additional parameters. Based on these data, a graph is created in which vertices represent individual news items, while edges describe the degree of their similarity. The Leiden method performs the step-by-step merging of such news items into clusters, ensuring high-quality partitioning due to iterative refinement of group structures. Using Leiden clusters makes it possible to identify concentrations of messages that exhibit signs of falsification or coordinated dissemination. Typically, such groups are characterized by repeated phrases, identical text templates, accelerated dissemination dynamics, and a large number of cross-references between non-authoritative sources. Analyzing these structures allows the detection of targeted information campaigns, bot networks, and mass distributions of false content. Thus, the Leiden clustering method is an effective approach for summarizing and grouping news messages, significantly improving the accuracy of fake-news detection. Its application in media monitoring systems provides deeper insight into information flows and contributes to the timely identification of disinformation.