Thirty years of public speaking anxiety research: topic modeling and semantic trend forecasting using LDA–Word2Vec integration

Ming C. Lin, Yi Sun, Lang ChenFrontiers in Human Neuroscience 2025 · DOI

Abstract

Public speaking anxiety (PSA) is a widespread condition with profound psychological, educational, and occupational consequences. Despite increasing academic attention, few studies have systematically mapped the thematic development of PSA research or anticipated future directions. This study employs a semantically enriched framework integrating latent Dirichlet allocation (LDA) with Word2Vec to analyze 990 peer-reviewed journal articles (1995-2024) retrieved from the Web of Science Core Collection. Nine major thematic clusters are identified, covering topics such as mental health distress, physiological reactivity, language apprehension, social phobia, and virtual reality interventions. Thematic evolution across four historical phases reveals growing conceptual convergence and an increasing focus on digital anxiety and assessment technologies. This study provides a comprehensive, data-driven mapping of PSA research and offers methodological guidance for future interdisciplinary bibliometric analyses.

Introduction

Public Speaking Anxiety (PSA)—often referred to as "glossophobia"—is one of the most prevalent social phobias, significantly impacting academic performance and career trajectories. While the volume of research has grown substantially over the last three decades, the landscape has remained fragmented across psychology, education, and communication sciences. Existing reviews are frequently qualitative or narrow in scope, leaving a gap in our understanding of the field's semantic evolution. This paper addresses this by providing a systematic, data-driven map of PSA research from 1995 to 2024, utilizing advanced natural language processing (NLP) to identify shifts and predict future research frontiers.

Methodology: The LDA–Word2Vec Framework

To move beyond simple citation counts, the researchers implemented a "semantically enriched" framework. This pipeline integrates Latent Dirichlet Allocation (LDA) for statistical topic identification with Word2Vec embeddings to capture deeper semantic relationships that traditional models often miss.

Methodological workflow of the study.

The methodology involved several sophisticated steps: identifying nine distinct probabilistic topics via LDA, training a Word2Vec model on the corpus to create "Topic Vectors," and using K-means clustering to observe how these topics grouped together over time. To predict future trends, the authors introduced two novelty indicators—Cluster Novelty and Topic Novelty—which calculate the average publication year of documents within a group to identify "fresher" research areas.

The analysis of 990 peer-reviewed articles from the Web of Science Core Collection reveals a clear upward trajectory in PSA research interest. The study identifies four distinct historical phases based on publication volume and thematic focus.

Annual publication trends in public speaking anxiety research (1995–2024).
  • Emergent (1995–2004): A period of low, exploratory output.
  • Early Expansion (2005–2013): Steady growth with a primary focus on clinical psychology.
  • Consolidation (2014–2019): Increased interdisciplinary integration.
  • High Productivity (2020–2024): A surge in research focusing on digital anxiety, online education, and the psychological impacts of the COVID-19 pandemic.

Key Thematic Insights

The field has transitioned from early fragmentation to modern conceptual convergence. While early research consisted of many small, disconnected clusters, the field has matured into fewer, more cohesive, and highly specialized areas.

A major finding is the rise of the "Digital Cure." Topic 6 (Virtual Exposure and Treatment Design) emerged as the most dominant theme overall, reflecting a shift from traditional "talk therapy" to technology-mediated interventions like Virtual Reality (VR). Additionally, the study noted a "COVID-19 Insertion"—a sudden, isolated theme regarding pandemic-related mental health that appeared in the 2020–2024 phase, demonstrating how global crises can force rapid pivots in research focus.

Future Frontiers

By applying novelty metrics, the study identified three primary "hotspots" that represent the current research frontier:

  1. Language Learning (Topic 4): Anxiety specifically related to non-native speakers and language acquisition.
  2. Virtual Therapy (Topic 6): Continued refinement of VR and digital exposure treatments.
  3. Psychometric Tools (Topic 8): A move toward more nuanced, empirically validated, and scalable measurement scales.

Conclusion and Limitations

While the LDA-Word2Vec integration offers a powerful tool for "horizon scanning," the study is limited by its focus on English-language articles and the inherent subjectivity involved in labeling data-driven topics. Future research should focus on cross-cultural scales, real-time monitoring methods (such as wearable bio-sensors), and AI-assisted training systems that adapt to individual speaker needs. This study provides a robust methodological foundation for future interdisciplinary bibliometric analyses.

Summary updated

Abstract

Public speaking anxiety (PSA) is a widespread condition with profound psychological, educational, and occupational consequences. Despite increasing academic attention, few studies have systematically mapped the thematic development of PSA research or anticipated future directions. This study employs a semantically enriched framework integrating latent Dirichlet allocation (LDA) with Word2Vec to analyze 990 peer-reviewed journal articles (1995-2024) retrieved from the Web of Science Core Collection, spanning SSCI, SCI-Expanded, ESCI, and A&HCI indices. Nine major thematic clusters are identified, covering topics such as mental health distress, physiological reactivity, language apprehension, social phobia, and virtual reality interventions. Thematic evolution across four historical phases reveals growing conceptual convergence and an increasing focus on digital anxiety and assessment technologies. A Sankey diagram illustrates patterns of thematic continuity and transformation, highlighting both enduring topics and emerging lines of inquiry. To predict future research frontiers, two novelty indicators-Cluster Novelty and Topic Novelty-are introduced. Clusters focused on language learning (Topic 4), virtual therapy (Topic 6), and psychometric tools (Topic 8) emerge as the most temporally novel. This study provides a comprehensive, data-driven mapping of PSA research and offers methodological guidance for future interdisciplinary bibliometric analyses.
BibTeX
@inproceedings{lin2025thirty,
  author = {Ming C. Lin and Yi Sun and Lang Chen},
  title = {{Thirty years of public speaking anxiety research: topic modeling and semantic trend forecasting using LDA–Word2Vec integration}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2025},
  doi = {10.3389/fnhum.2025.1653899},
}
Figure