Scholarlink

<p>In the ever-evolving realm of social media platforms, crucial insights into the mental well-being of users have surfaced, especially within the Bangla speaking community. Against the backdrop of a global mental health crisis, where approximately 21% of adults grapple with mental disorders, with over half remaining untreated, this study introduces an innovative method for identifying potential mental health disorder risks within Bengali social media content. Utilizing a carefully curated dataset of 7,131 Bengali expressions collected from various social media platforms and related to mental health, which has been validated by clinical experts, this research employs both traditional machine learning techniques and advanced deep learning methods. Importantly, this study acknowledges the challenges faced by individuals seeking help on social media, even when they may not be fully aware of their specific issues. Noteworthy is the proposal of a weighted ensemble of transformer methodologies, incorporating m-BERT, Bangla-BERT, and XLM-R as pivotal classifiers for the detection of mental health disorders in Bengali. This groundbreaking model evaluates the SoftMax probabilities of the classifiers based on their initial outputs. Through the utilization of this sophisticated weighting methodology, the model surpasses established machine learning and deep learning standards, achieving a remarkable weighted f1-score of 97% in the detection of mental health disorders. This study opens a new avenue for identifying mental health issues within Bengali social media, highlighting the importance of timely and essential interventions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

What’s on Your Mind? - Mental Health Disorder Risk Detection From Bangla Social Media Text Using Weighted Ensemble of Transformers

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Farzana Tasnim,
  • Sadia Sharmin

摘要

In the ever-evolving realm of social media platforms, crucial insights into the mental well-being of users have surfaced, especially within the Bangla speaking community. Against the backdrop of a global mental health crisis, where approximately 21% of adults grapple with mental disorders, with over half remaining untreated, this study introduces an innovative method for identifying potential mental health disorder risks within Bengali social media content. Utilizing a carefully curated dataset of 7,131 Bengali expressions collected from various social media platforms and related to mental health, which has been validated by clinical experts, this research employs both traditional machine learning techniques and advanced deep learning methods. Importantly, this study acknowledges the challenges faced by individuals seeking help on social media, even when they may not be fully aware of their specific issues. Noteworthy is the proposal of a weighted ensemble of transformer methodologies, incorporating m-BERT, Bangla-BERT, and XLM-R as pivotal classifiers for the detection of mental health disorders in Bengali. This groundbreaking model evaluates the SoftMax probabilities of the classifiers based on their initial outputs. Through the utilization of this sophisticated weighting methodology, the model surpasses established machine learning and deep learning standards, achieving a remarkable weighted f1-score of 97% in the detection of mental health disorders. This study opens a new avenue for identifying mental health issues within Bengali social media, highlighting the importance of timely and essential interventions.