Real Talk, Virtual Faces: A Formal Concept Analysis of Personality and Sentiment in Influencer Audiences
Summary: arXiv:2603.24410v1 Announce Type: cross
Abstract
Virtual influencers (VIs) — digitally synthetic social-media personas — are gaining traction in the digital landscape, attracting audiences whose discourse appears qualitatively different from that surrounding human influencers (HIs). Existing literature has attempted to characterize this difference primarily through surveys or aggregate engagement statistics, which elucidate what audiences communicate but not how diverse signals coalesce. To address this gap, we propose a two-layer, structure-first framework grounded in Formal Concept Analysis (FCA) and association rule mining.
Framework Overview
The proposed framework consists of two significant layers:
- Layer One: This layer employs FCA with support-based iceberg filtering to analyze weekly-aggregated comment data. It extracts discourse profiles that consist of weekly co-occurrence bundles encompassing sentiment, Big Five personality cues, and topic tags.
- Layer Two: This layer involves mining association rules at the comment level, unveiling personality-sentiment-topic dependencies that are often obscured in traditional frequency-table analyses.
Findings from Influencer Analysis
Our two-layer analysis has been applied to YouTube comments from three pairs of VIs and HIs. The results reveal a consistent structural divergence in discourse:
- Human influencer discourse tends to concentrate into a singular, emotionally regulated regime characterized by low neuroticism, which anchors a generally positive tone.
- In contrast, virtual influencer discourse reveals three structurally distinct modes of discourse, including a notable appearance-discourse cluster that is conspicuously absent from human influencer interactions, despite a near-equal marginal prevalence.
Topic-Specific Insights
Further analyses focusing on specific topics indicate that contexts related to virtual influencers tend to exhibit negative sentiment in psychologically sensitive domains such as:
- Mental health
- Body image
- Artificial identity
These domains show a stark contrast when compared to human influencer contexts, which generally maintain a more positive sentiment.
Conclusion
Our findings position Formal Concept Analysis as a robust tool for conducting multi-signal discourse analysis, illustrating that the phenomenon of virtuality significantly reshapes not only the content of audience discourse but also the fundamental grammar of how various signals co-occur in their reactions. This research opens new avenues for understanding the dynamics of influencer-audience interactions in an increasingly digital world.