MentalImager: Exploring Generative Images for Assisting Support-Seekers' Self-Disclosure in Online Mental Health Communities

TL;DR

Two user studies demonstrate that MentalImager not only improves seekers' satisfaction with their self-disclosure in their posts but also invokes support-providers' empathy for the seekers and willingness to offer help.

Abstract

Support-seekers' self-disclosure of their suffering experiences, thoughts, and feelings in the post can help them get needed peer support in online mental health communities (OMHCs). However, such mental health self-disclosure could be challenging. Images can facilitate the manifestation of relevant experiences and feelings in the text; yet, relevant images are not always available. In this paper, we present a technical prototype named MentalImager and validate in a human evaluation study that it can generate topical- and emotional-relevant images based on the seekers' drafted posts or specified keywords. Two user studies demonstrate that MentalImager not only improves seekers' satisfaction with their self-disclosure in their posts but also invokes support-providers' empathy for the seekers and willingness to offer help. Such improvements are credited to the generated images, which help seekers express their emotions and inspire them to add more details about their experiences and feelings. We report concerns on MentalImager and discuss insights for supporting self-disclosure in OMHCs.

Figures (8)

• Figure 1: Our studies to design and evaluate MentalImager for facilitating mental health self-disclosure with generated images in online communities.
• Figure 2: Screenshot (parts a-e) of the MentalImager embedded in a simulated community. (a) Area for creating a post. (b, b', and b") Buttons for detecting keywords (only appear in Study III), generating images, continuing to submit, and re-editing the post. (c) Keyword editor panel. (d and e) Panel for regenerating and displaying images. (f) Buttons for uploading an image and submitting the post in Study II's baseline interface. (g) Regenerated images based on the edited keywords.
• Figure 3: Four methods to generate images, which are: (a) Baseline: Extract keywords from the user text and input them into Google Images for retrieval, using the first downloadable image as the result. In the example, the output is an image of a man sitting on a bench (red box). (b) Content-based: Directly input the user text content into the Stable Diffusion (SD) model to generate an image. In the example, the output is a collection of sad face (orange box). (c) Emo-keyword-based (proposed): Input the extracted keywords along with additional topic and emotion keywords into the SD model to generate an image. In the example, the output is a girl in her room with many gaming devices (blue box). (d) Keyword-based: Input the extracted keywords into the SD model to generate an image. In the example, the output is a Netflix-style poster image (green box).
• Figure 4: Means and standard errors of the perceived visual quality of the images as well as their relevance to the input textual posts in an online mental health community. The images are either retrieved using the Baseline approach or generated using Content-based, Keyword-based, or Emo-keyword-based textual prompts. 5-point Likert scale; $+ : .05 < p < .1, * : p < .05, ** : p < .01, *** : p < 0.001$.
• Figure 5: Example post and its four related images in Study I. (a) Baseline. (b) Content-based. (c) Keyword-based: in this example, the keywords are "broken, happy, depressed, pity, displaying self-pity, 0 self-confidence, someone please get, said, works, understands". (d) Emo-keyword-based: in this example, apart from the keywords in (c), it additionally uses detected use the emotional keyword "sadness" and the keyword from the title "hopeless". We mask sensitive details of the example post for privacy concerns.