Podcast Outcasts: Understanding Rumble's Podcast Dynamics

TL;DR

This study addresses how podcast platforms influence political discourse by comparing YouTube and Rumble podcasts through large-scale audio-visual analysis. It combines speech-to-text transcription, BERTopic-based topic modeling with MPNet embeddings, and CLIP-based visual clustering to quantify political bias, engagement drivers, and visual strategies. The findings reveal a pronounced right-wing bias on Rumble, with topics, channels, and visuals aligning with right-wing content, while YouTube exhibits broader, more apolitical content; controversial topics and deplatforming events significantly boost Rumble views. The work highlights platform effects on polarization and suggests avenues for cross-platform analysis and algorithmic investigation of content amplification, with implications for researchers and policymakers studying online political communication.

Abstract

Podcasting on Rumble, an alternative video-sharing platform, attracts controversial figures known for spreading divisive and often misleading content, which sharply contrasts with YouTube's more regulated environment. Motivated by the growing impact of podcasts on political discourse, as seen with figures like Joe Rogan and Andrew Tate, this paper explores the political biases and content strategies used by these platforms. In this paper, we conduct a comprehensive analysis of over 13K podcast videos from both YouTube and Rumble, focusing on their political content and the dynamics of their audiences. Using advanced speech-to-text transcription, topic modeling, and contrastive learning techniques, we explore three critical aspects: the presence of political bias in podcast channels, the nature of content that drives podcast views, and the usage of visual elements in these podcasts. Our findings reveal a distinct right-wing orientation in Rumble's podcasts, contrasting with YouTube's more diverse and apolitical content.

Figures (6)

• Figure 1: Heatmaps illustrating the cosine similarities among the top N topic centroids: (a) YouTube versus Rumble, and across left-wing, center, and right-wing podcasts; (b) Rumble versus YouTube, and across left-wing, center, and right-wing podcasts. Darker shades denote greater semantic similarity.
• Figure 2: Density plots of political alignment scores for Rumble and YouTube channels. Scores represent ideological orientation and range from -1 to 1, where negative values denote a left-leaning bias and positive values suggest a right-leaning inclination.
• Figure 3: Scatter plot showing right-wing and left-wing similarity distributions for the top 320 topics in podcast videos of popular YouTube and Rumble podcast channels, with R-squared and slope values from linear regression.
• Figure 4: CDF of the proportion of sentences covered cumulatively at each topic rank in YouTube and Rumble podcast videos. Topic ranks start at 20 and increase exponentially.
• Figure 5: Comparison of visual topics between Youtube and Rumble, extracted through clustering, showing top-10 clusters for each platform (Refer to Table \ref{['tab:comparison_top_images']}).