Social Network Analysis on Audio-Based Platforms

🔬 Ongoing Research

The rise of audio-based platforms such as X Spaces (formerly Twitter Spaces) introduces synchronous, ephemeral modes of interaction that differ fundamentally from traditional text-based social media. This research investigates user connectivity and thematic engagement by analysing large-scale datasets of X Spaces events.

Unlike text-based social networks where content persists and can be analysed retrospectively, audio spaces are live, temporary events that create unique challenges and opportunities for understanding how users express interests and form communities.

Research Questions

How do users connect and form communities through audio-based interactions?
What structural patterns emerge in user–Space participation networks?
How do user interests align (or diverge) across audio and text modalities?
Do live audio conversations stay on topic or drift from their stated themes?

Technical Approach

Network Analysis

User participation is modelled as a bipartite user–Space network, then projected into user–user connections to assess community structure, interaction patterns, and user influence.

The research employs multiple analytical methods:

Network Analysis: Examining broad-scale characteristics and local structures
Topic Modelling: Extracting conversation topics from transcribed audio summaries
Cross-Modal Comparison: Comparing creator-assigned topics, conversation-derived topics, and users' text posts
Hybrid Alignment Method: Combining BERT embeddings, spaCy similarity, and expert validation

Key Findings

Weak Cross-Modal Alignment

User interests are only weakly aligned across audio and text modalities, indicating distinct communicative roles for each medium.

Topic Drift

While most Spaces exhibit high topic coverage, 44% introduce additional themes beyond their predefined scope.

Network Structure

Analysis confirms broad-scale network characteristics while providing novel insights into local structures and user influence.

Distinct Roles

Audio and text platforms serve different communicative purposes, suggesting users express different aspects of their interests through each.

Methodology Contributions

Hybrid Topic Alignment: Novel method combining BERT embeddings with spaCy similarity and expert validation
Bipartite Network Projection: Approach for modelling participation in ephemeral audio events
Cross-Modal Interest Analysis: Framework for comparing user behaviour across different content modalities

Implications

This research contributes to understanding interest expression, structural cohesion, and topical drift in emerging audio-based social platforms. The findings have implications for:

Platform design for audio-based social features
Content recommendation across modalities
Understanding how users build communities through synchronous audio
Multimodal user profiling approaches