Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM Use for Mental Health: Crowdsourcing Users' Sentiment-based Perspectives and Values from Social Discussions

Lingyao Li, Xiaoshan Huang, Renkai Ma, Ben Zefeng Zhang, Haolun Wu, Fan Yang, Chen Chen

TL;DR

Using a Value-Sensitive Design framework, this study crowdsources posts from six social platforms to examine how users discuss interactions with LLM chatbots for mental health across multiple conditions. It applies an LLM-assisted extraction pipeline to label sentiment, perspectives, and values, yielding condition-specific patterns and an impact typology. The findings show neurodivergent groups report positive sentiment and instrumental support, while higher risk disorders show negative sentiment and safety concerns, highlighting the need for condition-specific, value-sensitive LLM design. The work provides design implications to balance user autonomy, privacy, and safety in real world mental health AI deployments.

Abstract

Large language models (LLMs) chatbots like ChatGPT are increasingly used for mental health support. They offer accessible, therapeutic support but also raise concerns about misinformation, over-reliance, and risks in high-stakes contexts of mental health. We crowdsource large-scale users' posts from six major social media platforms to examine how people discuss their interactions with LLM chatbots across different mental health conditions. Through an LLM-assisted pipeline grounded in Value-Sensitive Design (VSD), we mapped the relationships across user-reported sentiments, mental health conditions, perspectives, and values. Our results reveal that the use of LLM chatbots is condition-specific. Users with neurodivergent conditions (e.g., ADHD, ASD) report strong positive sentiments and instrumental or appraisal support, whereas higher-risk disorders (e.g., schizophrenia, bipolar disorder) show more negative sentiments. We further uncover how user perspectives co-occur with underlying values, such as identity, autonomy, and privacy. Finally, we discuss shifting from "one-size-fits-all" chatbot design toward condition-specific, value-sensitive LLM design.

LLM Use for Mental Health: Crowdsourcing Users' Sentiment-based Perspectives and Values from Social Discussions

TL;DR

Using a Value-Sensitive Design framework, this study crowdsources posts from six social platforms to examine how users discuss interactions with LLM chatbots for mental health across multiple conditions. It applies an LLM-assisted extraction pipeline to label sentiment, perspectives, and values, yielding condition-specific patterns and an impact typology. The findings show neurodivergent groups report positive sentiment and instrumental support, while higher risk disorders show negative sentiment and safety concerns, highlighting the need for condition-specific, value-sensitive LLM design. The work provides design implications to balance user autonomy, privacy, and safety in real world mental health AI deployments.

Abstract

Large language models (LLMs) chatbots like ChatGPT are increasingly used for mental health support. They offer accessible, therapeutic support but also raise concerns about misinformation, over-reliance, and risks in high-stakes contexts of mental health. We crowdsource large-scale users' posts from six major social media platforms to examine how people discuss their interactions with LLM chatbots across different mental health conditions. Through an LLM-assisted pipeline grounded in Value-Sensitive Design (VSD), we mapped the relationships across user-reported sentiments, mental health conditions, perspectives, and values. Our results reveal that the use of LLM chatbots is condition-specific. Users with neurodivergent conditions (e.g., ADHD, ASD) report strong positive sentiments and instrumental or appraisal support, whereas higher-risk disorders (e.g., schizophrenia, bipolar disorder) show more negative sentiments. We further uncover how user perspectives co-occur with underlying values, such as identity, autonomy, and privacy. Finally, we discuss shifting from "one-size-fits-all" chatbot design toward condition-specific, value-sensitive LLM design.

Paper Structure

This paper contains 24 sections, 2 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs and Mental Health
  4. Achieving Value-Sensitivity
  5. Methods
  6. Data Preparation
  7. Conceptual Design
  8. Information Extraction
  9. Performance Validation
  10. User Perspective Standardization
  11. Results
  12. Sentiment across mental conditions in LLM
  13. Perspectives and Values for Mental Health
  14. Value-Perspective Co-Occurrences in User Sentiment
  15. Mental Health Condition-Specific Sentiment, Values, Perspective
  16. ...and 9 more sections

Figures (5)

  • Figure 1: Research framework to implement data preparation and methods.
  • Figure 2: Time-series trends: (a) Monthly post volume, (b) Average sentiment over time, (c) Post volume by mental health condition, (d) Average sentiment by mental health condition, and (e) Fingerprint of LLM chatbot across mental health conditions.
  • Figure 3: Average sentiment variation across: (a) LLM chatbots. (b) Mental health conditions.
  • Figure 4:
  • Figure 5: Heatmap on the strength of associations between mental health (MH) conditions and user interaction variables based on Adjusted Standardized Residuals (ASR). (a) ASR of MH Condition vs. Sentiment. (b) ASR of MH Condition vs. Perspective. (c) ASR of MH Condition vs. Value. Abbreviations: Values (V): V_Ac: Accountability; V_Au: Autonomy; V_Ca: Calmness; V_ES: Environmental Sustainability; V_Ffb: Freedom From Bias; V_HW: Human Welfare; V_IC: Informed Consent; V_Id: Identity; V_OP: Ownership and Property; V_Pr: Privacy; V_Tr: Trust; V_UU: Universal Usability. Perspectives (P): P_An: Anthropomorphism; P_AS: Appraisal Support; P_CS: Clinical Skepticism; P_De: Dependency; P_ES: Emotional Support; P_Et: Ethics; P_InfS: Informational Support; P_InsS: Instrumental Support; P_IntL: Interaction Limitations; P_MU: Maladaptive Usage; P_PH: Psychological Harm; P_SI: Sociocultural Impact.