The Anatomy of Disinformation Networks: A Hybrid Graph-Based Framework for Echo Chamber Detection in the Italian Twitter/X Sphere

The proliferation of disinformation and the emergence of echo chambers on social media pose serious challenges for democratic discourse. In this paper, we introduce a hybrid computational framework that fuses network topology with user-generated hashtag semantics to map and measure candidate echo chamber structures at scale. Using the full Italian Twitter Firehose from February 23 to August 31, 2022, we built a heterogeneous graph of user and hashtag nodes linked by Twitter interactions. We then ran the Leiden algorithm over ten resolution settings and evaluated community quality via normalized mutual information (NMI), variation of information (VI), adjusted Rand index (ARI), coverage, and the external--internal index (EI). Our analysis reveals that the community structure is remarkably stable between resolution values of 0.5 and 0.6, where normalized mutual information approaches 0.999, the adjusted Rand index climbs to 0.89, and the variation of information stays below 0.05, marking this interval as optimal for coherent, semantically aligned groups. Collapsing all clusters smaller than 30 nodes distills the network into 1000 robust communities that still cover more than 54\% of the total edge weight and maintain a median size of 42–45 users while further enhancing ARI and VI. Finally, a second-level Leiden pass on the top 500 communities drives the EI index down to 0.0411, indicating maximal intragroup connectivity and the strongest echo chamber effects. Together, these results show that our dual structural–semantic framework not only uncovers stable, multilevel community hierarchies but also indicates the precise scale at which echo chambers intensify, paving the way for real-time disinformation monitoring and targeted mitigation.