User-Reported Issues with Mental Health Apps: A Machine-Assisted Topic Analysis of Social Media Posts (Preprint)
Bibliographic record
Abstract
Background: Mobile apps marketed to support mental health have become increasingly popular in recent years. Given their widespread use, it is important to identify issues that users experience while using such apps. Understanding these issues may provide insight into the safety and suitability of these apps for individuals seeking mental health support. Objective: Unlike existing research, where user experience issues have been identified through researchers' direct analysis of apps, this study aimed to generate themes relating to user experience issues using comments from app users themselves. An additional aim was to evaluate a human-in-the-loop machine learning approach using structural topic modeling (STM) to analyze vast volumes of data gathered from X (formerly Twitter, developed by Twitter, Inc). Methods: Data relating to five of the most popular mental health apps were collected from the X API using R. A machine-assisted thematic analysis approach combined STM with human qualitative analysis to interpret user-generated posts. An unsupervised topic-modeling approach was tested using models with 5-40 topics and differing covariates (ultimately, a model without covariates was selected). Two researchers independently conducted thematic analysis to interpret and contextualize model outputs. A structural topic model with 10 topics, each comprising 20 X posts, was selected as most appropriate for generating insights. Results: Using R (developed by the R Core Team), 79,703 X posts were collected via the X API relating to five popular mental health apps. After negative sentiment filtering, 19,603 posts remained. Posts spanned March 2006 (the launch of X/formerly Twitter) to December 2022. Researchers collaboratively labeled the 10 topics to identify the primary user experience issue represented in each. Topic 3 was discarded due to low coherence and inconsistency in relation to app user experience, and Topic 5 was discarded because posts reflected app X account activity rather than user experience of the apps. The remaining eight topics were organized into four themes. The first theme, guidance shortfall, included difficulties following guided meditations, challenges selecting appropriate content from large libraries, and incompatibility between app use and home environments. The second theme, technical difficulties, involved subscription access issues and technical faults within apps. The third theme, heightened emotions related to app-affiliated celebrities, captured both over-excitement linked to celebrity involvement and anger directed toward specific celebrities. The final standalone theme, negative impacts of sleep self-monitoring, demonstrated users reporting that tracking sleep adversely affected sleep experience. Conclusions: The combination of STM and human qualitative analysis of X posts identified several user-experienced issues associated with popular mental health apps, often linked to negative outcomes. This study provides evidence that STM can be combined with qualitative methods to rapidly analyze large-scale social media data and generate insights into user experience of mass-reach digital health interventions.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.011 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.000 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".