Members of NewsVibe’s team have authored a paper which will showcase the platform’s capabilities at a prestigious international academic forum later this month. Their study on Romanian NATO discourse has been accepted for presentation at Sunbelt 2025 – the annual International Network for Social Network Analysis (INSNA) conference, taking place at Sorbonne University in Paris (June 23–29, 2025). The paper demonstrates how NewsVibe’s media intelligence platform mapped shifts in Romanian news and Facebook discussions about NATO before and after the country’s controversial 2024 presidential election.
The paper, titled “Unveiling the dynamics of media discourse on NATO in Romania amid electoral turmoil through semantic network analysis”, was authored by Dr. Dan Sultanescu, Dr. Stefan Ruseti, Dr. Dana Sultanescu and Dr. Itai Himelboim, who served as academic advisor.
Dr. Himelboim (University of Georgia, US), a leading scholar of social network analysis, known for his work on media ecosystems, political communication and the structure of digital public discourse, advised the NewsVibe team throughout the research process.
Key findings
The semantic network analysis uncovered several notable shifts from pre-election to post-election discourse. Key findings include:
-
NATO mentions doubled after the election (rising from about 3,200 to 6,662 total references across news websites and Facebook)
-
The web of topics expanded and overlapped more. The network of key entities nearly doubled in size, and narrative clusters became less distinct, indicating that previously separate storylines began merging.
-
Anti-NATO voices and narratives gained visibility, shifting from isolated or diffuse criticism to a more focused, actor-centered opposition.
The post-election network (right) is visibly larger and more interconnected, reflecting the surge in discussion and the blending of previously separate narrative communities.
NewsVibe’s technical edge delivers scalable narrative mapping
Behind these insights is NewsVibe’s powerful media intelligence platform. For data collection, the system monitors over 7,500 online sources – including news sites, public Facebook pages, and YouTube channels – across Europe and the US, refreshing every 10 minutes to stay up-to-date.
It also employs advanced natural language processing: a fine-tuned named entity recognition (NER) model built on Llama AI that automatically identifies people, organizations, locations, brands, and other entities with high accuracy (around 82% F1 score in testing). These capabilities allow NewsVibe to detect subtle changes in narratives in real time and across platforms, providing a uniquely holistic view of the media landscape.
For this paper, data harvesting, named-entity recognition and co-occurrence matrix generation were performed using NewsVibe, while community detection and visualization were conducted in Gephi. NewsVibe has since integrated capabilities for automatic network clustering and visual mapping of relationships, enabling end-to-end semantic network analysis.
