Write a comment

This preprint uses BERTopic dynamic topic modelling on 7,254 Web of Science journal articles (2010–Jan 2025) to map how the research conversation on social media and well-being has evolved over time. It identifies 110 topics and argues that the field contains two main thematic “streams”: one centred on social connection and support, and another centred on anxiety-related outcomes such as social comparison, FoMO, body image concerns, and problematic use, with methodological/contextual themes (e.g., COVID-19 communication, AI-based detection) forming a third area.

Its main finding is a temporal shift in scholarly attention: work from roughly 2010–2016 is presented as relatively balanced between connection-orientated and anxiety-orientated frames, while research after about 2017 increasingly concentrates on anxiety-related themes, particularly FoMO and body image, with COVID-19 creating a sharp but time-bounded surge in related discourse. The paper moves the field forward by offering a data-driven, longitudinal “map” of how the literature’s topical priorities have changed, complementing existing meta-analyses that focus on effect sizes by instead characterising the evolution of themes and research emphases across time.

Major issues

• Potential bias in literature selection due to the search strategy requiring social media terms in titles and mental health constructs in the topic field, which may over-represent anxiety-focused research and under-represent broader or neutral perspectives

• Insufficient methodological transparency in the BERTopic implementation, with missing details on key parameters (UMAP, HDBSCAN, vectorisation, random seeds), limiting reproducibility and evaluation of robustness

• Absence of robustness and validation checks for the topic model, including topic stability, coherence, or sensitivity analyses to alternative model specifications

• Turning point and dominance claims are not statistically tested, relying on visual inspection of trends rather than formal change-point analysis or uncertainty estimation

• Interpretive overreach from discourse analysis to social reality, where shifts in academic attention are sometimes framed as direct reflections of platform-driven or societal psychological change without external validation

• Unclear topic labelling and construct validity, as topic names imply specific psychological mechanisms without explanation of how labels were assigned or validated against representative documents

Minor issues

• The title contains a typographical error (“Form” instead of “From”), which should be corrected for clarity and professionalism.

• The stated temporal scope of the dataset is slightly inconsistent across sections (e.g., “2010–2025” vs. “January 2010 to January 2025”), which may confuse readers.

• Several references use Google search links instead of direct DOIs or publisher URLs, and overall reference formatting is inconsistent.

• Encoding artefacts and minor formatting issues appear in the text, which interrupt readability and should be cleaned up.

• The process used to assign topic labels is not briefly explained, which would help readers better understand the interpretation of results.

• Figures and tables would benefit from slightly more descriptive captions to ensure they are fully interpretable on their own.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.