Article: Automated assessment of mental health integrating emojis and BERT Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.17 pp.1 - 14 Abstract: With social media evolving into a key platform for the public to express emotions and mental states, research on automated mental health assessment techniques has attracted growing attention. However, traditional methods suffer from drawbacks such as low efficiency. To address this, this study proposes a multi-level automated assessment model integrating emojis and bidirectional Transformer architecture. A text-emoji fuser with a dual attention mechanism is constructed, combined with a multimodal mental health lexicon and bidirectional long short-term memory network to realise end-to-end mental state analysis. Experimental results show the model achieves 0.968 macro-AUC and 89.52% accuracy in four-class risk assessment, and retains 0.865 macro-F1 in sentiment-conflicting scenarios. Ablation experiments verify the validity of core modules, with the fusion module reaching 94.16% accuracy and 92.87% F1 score. This model enables efficient assessment, provides reliable technical support for large-scale screening systems, and contributes significantly to the intelligent development of public mental health services. Inderscience Publishers - linking academia, business and industry through research

Title: Automated assessment of mental health integrating emojis and BERT

Authors: Bin Song

Addresses: Development Planning Office, Shenyang Polytechnic College, Shenyang, 1100045, China

Abstract: With social media evolving into a key platform for the public to express emotions and mental states, research on automated mental health assessment techniques has attracted growing attention. However, traditional methods suffer from drawbacks such as low efficiency. To address this, this study proposes a multi-level automated assessment model integrating emojis and bidirectional Transformer architecture. A text-emoji fuser with a dual attention mechanism is constructed, combined with a multimodal mental health lexicon and bidirectional long short-term memory network to realise end-to-end mental state analysis. Experimental results show the model achieves 0.968 macro-AUC and 89.52% accuracy in four-class risk assessment, and retains 0.865 macro-F1 in sentiment-conflicting scenarios. Ablation experiments verify the validity of core modules, with the fusion module reaching 94.16% accuracy and 92.87% F1 score. This model enables efficient assessment, provides reliable technical support for large-scale screening systems, and contributes significantly to the intelligent development of public mental health services.

Keywords: emoji-text fusion; bidirectional encoder representations from transformers; BERT; multimodal fusion; automated mental health assessment; bidirectional long short-term memory network; BiLSTM.

DOI: 10.1504/IJRIS.2026.154535

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.17, pp.1 - 14

Received: 22 Dec 2025
Accepted: 01 May 2026
Published online: 02 Jul 2026 *

Title: Automated assessment of mental health integrating emojis and BERT

Keep up-to-date