MIRoBERTa: Mental Illness Text Classification With Transfer Learning on Subreddits

This paper proposes MIRoBERTa, a domain-adaptive RoBERTa model pretrained on Reddit mental health content to enhance multiclass text classification. The model outperforms other traditional, deep learning, and transformer baselines, achieving an F1-score of 0.847. The study also explores ensemble techniques and word importance to improve interpretability and prediction performance.

Recommended citation: M. Sao and H.-J. Lim, "MIRoBERTa: Mental Illness Text Classification With Transfer Learning on Subreddits," IEEE Access, vol. 12, pp. 197454–197466, Dec. 2024.
Download Paper