Exploring mental health literacy on twitter: A machine learning approach

Objectives: This study investigates whether reducing mental illness stigma, enhancing help-seeking efficacy, and maintaining positive mental health mediate the relationship between the recognition of mental disorders and help-seeking attitudes. Methods: During annotation phase, Twitter were collected data from April to August 2022. Tweets were retrieved using keywords aligned with five mental health literacy (MHL) facets: maintaining positive mental health (M), recognizing mental disorders (R), reducing mental illness stigma (S), help-seeking attitude (HA), and help-seeking efficacy (HE). A pretrained Sentence-BERT model generated embedding vectors for classification tasks, achieving 0.85 precision and 0.88 accuracy. Tweets from November 2021 to December 2022 were organized into three time points: R at Time 1; M, S, and HE at Time 2; and HA at Time 3. In total, 4,471,951 tweets from 941 users were analyzed. Structural equation modeling was employed to examine the temporal relationships among MHL components. Results: Single mediation models indicated that better recognition of mental disorders is associated with more favorable maintenance of positive mental health, greater help-seeking efficacy, and lower mental illness stigma—all of linked to more positive help-seeking attitudes. However, in the multiple mediation model, the reduction of mental illness stigma did not significantly mediate the relationship between the recognition of mental disorders and help-seeking attitudes. Conclusions: This findings suggest that recognizing mental disorders influences help-seeking attitudes through mediators like help-seeking efficacy and positive mental health maintenance. These results provide valuable insights for future interventions and policies aimed at promoting help-seeking behaviors and advancing mental health literacy.