Deep Representation of Discursive Evolution via Semantic Embedding of Virtual Identities in Online Marginalized Communities

This paper proposes a deep semantic embedding frame to computationally model the discourse evolution of virtual identities in online marginal communities, aiming to explore how marginal groups gain discourse power and expression space online through virtual identities.This experiment takes the datasets of the LGBTQ+ community on Reddit and Douban as examples, constructing a multi-dimensional model of user-speech-time to track the evolving expression paths of virtual identities, thereby validating the models ability to detect shifting discourse dominance and predict the dynamics of semantic adaptation within marginalized groups after the strategic adjustment. Findings reveal that marginalized users actively reshape mainstream discourse through linguistic strategies such as semantic ambiguity and emotional moderation, significantly improving their visibility and influence. This study challenges traditional views of the dynamic relationship between virtual identities and discourse power and expands the understanding of the online expression behaviors of marginalized groups. Practically, we provide new ideas for social media platforms to protect the voices of marginalized groups and call for greater attention to their voices.