JMIR Publications

ABSTRACT

Background:

Systemic lupus erythematosus (SLE) is a multifactorial autoimmune disease influenced by genetic, epigenetic, ecological, and environmental factors. Social media has become an important channel for the public to express health concerns and gain support. Reddit and Weibo, as mainstream social platforms both globally and in China, gather a massive amount of user generated content, providing valuable data sources for studying patients' real experiences.

Objective:

This study aimed to apply NLP techniques to analyze Reddit and Weibo posts related to SLE, exploring their temporal dynamics, and conducting sentiment analysis.

Methods:

This qualitative study examined all discussions related to SLE on social media platforms up to December 31, 2023. An artificial intelligence (AI)–driven analytical pipeline was established to categorize these discussions into distinct topics and overarching thematic domains. The pipeline integrated a semisupervised natural language processing framework based on the Bidirectional Encoder Representations from Transformers (BERT) model, coupled with dimensionality reduction and clustering algorithms. Sentiment polarity for each discussion was subsequently classified as positive, neutral, or negative using a pretrained BERT-based sentiment analysis model.

Results:

On Reddit, 11,318 discussions pertaining to SLE were retrieved, contributed by 3,649 unique authors. The average number of characters per discussion was 535.51 (SD = 760.83). 92.55% of authors contributed between one and five discussions. The most prevalent topics included discussions of SLE symptoms, expressions of gratitude and emotional support, and experiences related to malar rash diagnosis, skin biopsy, and clinical visits. While on Weibo, a total of 33,628 SLE related discussions were collected from 21,509 unique authors. The average number of characters per discussion was 265.38 (SD = 440.18), with 97.75% of authors contributing between one and five discussions. 99 and 189 SLE related topics were identified from the Reddit and Weibo datasets, respectively. The most prominent topics focused on breakthroughs in the use of artemisinin for SLE treatment, family caregiving and emotional struggles, and daily life experiences and emotional states of SLE patients. The overall sentiment of 11,318 SLE-related discussions on Reddit was positive, with a mean sentiment score of 0.42 (SD = 0.86). On Weibo, the overall sentiment of 33,628 discussions was neutral, with a mean sentiment score of 0.01 (SD = 0.91).

Conclusions:

This study demonstrates the feasibility of using artificial intelligence to perform topic modeling and sentiment analysis on large-scale, cross-cultural, and cross-platform social media data related to SLE.