# XLM-RoBERTa with Multi-View Representation Pooling for Mental Health Risk Classification in Indonesian Social Media Posts > Utami N.F. URL kanonis: https://discover.unhas.ac.id/publications/pub_scopus_105055423599 Jurnal / Konferensi: 2026 IEEE International Conference on Artificial Intelligence and Mechatronics Systems Aims 2026 Tahun terbit: 2026 DOI: https://doi.org/10.1109/AIMS71270.2026.11712858 Citations: 0 ## Authors - Utami N.F. ## Abstract Expressions related to anxiety and depression are increasingly being found on Indonesian social media. However, automatically classifying these expressions is not straightforward because social media posts often use informal, ambiguous, and code-mixed language with subtle emotional cues. To address this issue, this study introduces XLMR+MVRP, an XLM-RoBERTa- based model with Multi-View Representation Pooling, to classify posts into Normal, Anxiety-risk, and Depression-risk categories. The model combines weighted layer pooling with CLS, mean, max, and attention-based pooling to capture different sentence-level features. It was evaluated on 10,001 public X posts that were manually labeled into three categories. A total of 100 samples were reviewed by a psychologist, resulting in Cohen’s kappa score of 0.652. Because the psychologist had limited time to review the data, the 100 reviewed samples were treated as an initial reliability check, not as a validation of the full dataset. In the experiments, XLMR+MVRP yielded the strongest results among the tested models, achieving an accuracy of 0.9140 and a macroF1 score of 0.9142. The performance improvement over the two baseline models was statistically significant, as demonstrated by the paired bootstrap confidence intervals and McNemar test. This model was designed for screeningoriented text classification, not for clinical diagnosis. ## Keywords - Indonesian - Social media - Psychology - Mental health - Representation (politics) - Pooling - Social psychology - Social representation - Applied psychology - Computer science - Style (visual arts) - Risk perception - Identity (music) - Artificial intelligence --- Sumber: Discover Unhas — RIMS Universitas Hasanuddin. Saat mengutip, gunakan DOI bila tersedia atau URL kanonis di atas.