Data Visualization
Propagation Path
In this study, we employed Python’s Streamlit library to develop an interactive dashboard for analyzing the dissemination patterns and content characteristics of online rumors. The system specifically visualizes propagation pathways, classifies textual content, and performs sentiment analysis. For a selected rumor case, we first constructed a Propagation Path Dynamic Graph in Figure 1 to map its diffusion trajectory.
The visualization revealed multiple distinct central nodes representing high-impact users who acted as critical amplifiers in the rumor-spreading network. These key accounts exhibited a radial dissemination pattern, demonstrating how the rumor cascaded through hierarchical layers of users before reaching peripheral audiences. The graph structure confirmed a multi-level dissemination mechanism, where the rumor transitioned from core propagators to secondary sharers before achieving viral spread. To complement the structural analysis, the dashboard integrates text mining modules that categorize posts by thematic clusters and apply sentiment analysis to quantify emotional polarity across propagation tiers. This dual approach—combining network topology with content semantics—enables researchers to simultaneously identify superspreader accounts, trace contamination paths, and assess how emotional valence evolves during rumor diffusion. The Streamlit interface further allows real-time filtering by node centrality metrics, propagation depth, and sentiment thresholds, offering an integrated analytical framework for misinformation studies.
To contrast the dissemination dynamics, we analyzed a non-hoax COVID-19 post in Figure 2, focusing on debunking misinformation and disseminating scientific knowledge.
The original post primarily contained factual corrections and public health education, differing significantly from the emotional or sensational narratives observed in hoaxes. Visualization via the Streamlit dashboard revealed distinct propagation patterns: the fact-checking post exhibited a single-core diffusion structure, with most shares originating directly from authoritative institutional accounts. In contrast, the previously examined rumor demonstrated multi-hub propagation, characterized by numerous influential nodes independently amplifying the content. This structural divergence suggests that rumors leverage decentralized networks to achieve faster and broader dissemination, whereas fact-based content relies on centralized credibility for a slower but more controlled spread. The comparison underscores how network topology and content type jointly shape information virality.
Word Cloud
A word cloud visualization shown in Figure 3 of the other category rumor’s textual content, revealed a high concentration of negative adjectives such as “terrifying,” “horrific,” “disgusting,” and “absurd.” These lexically charged terms demonstrate the rumor’s reliance on fear-inducing and inflammatory language, which aligns with sentiment analysis results showing dominant negative polarity. The prevalence of emotionally loaded vocabulary—particularly those evoking fear (“terrifying”), disgust (“disgusting”), and outrage (“absurd”)—suggests intentional or emergent amplification of affective states among recipients. The stark divergence in the affective lexicon further explains the rumor’s rapid dissemination, as heightened emotional arousal often correlates with increased sharing behavior in social networks.
The word cloud in Figure 4 generated from the COVID-19 non-rumor post demonstrates its primary focus on fact-checking and science communication.
