Explainable Self-Attentive Transformer Model for Bangla Mental Health Disorder Detection

DOI

10.2991/978-94-6239-664-7_28How to use a DOI?

Keywords

Bangla mental health detection; Transformer model; Self-attention; Social media text analysis; Explainable AI

Abstract

Detecting mental health expressions in Bangla social media text remains a critical challenge, particularly in a rapidly digitalizing society where users increasingly express emotions and psychological distress online. We used the B-MHD (Bangla Mental Health Disorder Text) dataset, a manually annotated collection of 7,130 Bangla and BanglaEnglish code-mixed social media posts gathered from Facebook, YouTube, Twitter, and Reddit. While previous studies have explored traditional and transformer-based approaches for Bangla sentiment and depression detection, the integration of explainability into transformer architectures for mental health analysis remains underexplored. To address this gap, we conduct a systematic evaluation of classical, recurrent, and transformer-based models for Bangla mental health detection. As part of this evaluation, we employ various machine learning and deep learning models including SVM, Logistic Regression, BiLSTM, GRU, and LIME-based explanation analysis to investigate token-level contributions and model behaviour. Building upon these insights, this paper introduces one of the first explainable self-attentive transformer-based models designed specifically for Bangla mental-health text recognition, incorporating an attention and mean-pooling mechanism to enhance contextual understanding and interpretability. Experimental findings demonstrate that transformer-based models outperform traditional methods, with BanglaBERT combined with self-attention pooling achieving an F1-score of 97.04% and an accuracy of 96.98%. Through LIME-based explanation analysis, we further interpret token-level contributions, showing that emotionally expressive words strongly influence predictions, while metaphorical or contextually ambiguous phrases remain challenging. This study advances the development of explainable mental health detection systems for Bangla social media contexts.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Abu Saim Hossen Hridoy
AU  - Nazmus Sakib Shohan
AU  - Md. Shaharia Alif
AU  - Nurul Mursalin Ag Mahin
AU  - Md. Ayon Mia
PY  - 2026
DA  - 2026/06/08
TI  - Explainable Self-Attentive Transformer Model for Bangla Mental Health Disorder Detection
BT  - Proceedings of the International Conference on Intelligent Data Analysis and Applications (IDAA 2025)
PB  - Atlantis Press
SP  - 397
EP  - 408
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-664-7_28
DO  - 10.2991/978-94-6239-664-7_28
ID  - Hridoy2026
ER  -