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

This article focuses on developing automated methods for detecting fake news by analyzing graph structures and applying community detection algorithms in social networks. The rapid growth of digital news content makes it increasingly difficult to verify information reliability in real time, which drives the relevance of this study. Traditional approaches that rely solely on linguistic text analysis and classical machine learning are insufficient for identifying coordinated disinformation campaigns due to their network‑driven distribution patterns.

The authors propose shifting from isolated news analysis to a structural examination of information flows using a graph‑based model. In this model, individual news messages and information entities are represented as vertices, while their relationships form the edges of the graph. The primary focus is on applying Louvain and Leiden community detection algorithms, which cluster graphs by optimizing modularity and reveal densely connected message groups.

The article presents the architecture of a software tool designed for fake news detection, incorporating stages for data loading, preprocessing, graph construction, clustering, result analysis, and visualization. This tool follows a modular design, ensuring flexibility, scalability, and the ability to integrate different clustering algorithms as needed.

An experimental study compares the performance of the Louvain and Leiden algorithms using real‑world social network data. The results indicate that while both algorithms generate high‑purity communities, the Leiden algorithm produces more stable and internally cohesive clusters, thereby improving result interpretability and disinformation detection accuracy. Combining text similarity analysis with graph‑based community detection proves more effective than content‑only methods in uncovering fake news and orchestrated information campaigns.

The practical contribution of this work is a software tool with a graphical user interface that enables transparent, reproducible, real‑time analysis of news messages. This approach can support disinformation monitoring systems, inform information security decision‑making, and serve as a foundation for future research using dynamic graphs, multimodal data, and explainable artificial intelligence.

Author(s): Oksana Tsukan, Advisor to the Rector of Kharkiv National University of Internal Affairs