Hybrid Semantic and Graph-Based Analysis for News Consistency Detection

Abstract

The rapid propagation of online news, while facilitating access to information, has given rise to problems of misinformation and content inconsistency across different sources. This study proposes an AI-based system that automatically evaluates the contextual consistency and reliability of different online news articles reporting on the same event. The system collects news via web scraping, groups them by event, and applies preprocessing and extractive summarization. The summaries are encoded with SBERT, modeled as weighted graphs, and analyzed using PageRank and Z -score to compute a hybrid ConsistencyScore. Furthermore, the proposed hybrid method is applied to all core models used in the study, and the baseline and hybrid versions of each model are compared. The hybrid usage of the model yielded the highest accuracy, 94.55 %. This study aims to systematically demonstrate the impact of widely used embedding- and graphbased models by integrating them into an explainable, weighted system architecture specifically tailored to the problem of news consistency. © 2026 IEEE.