| G. Vonitsanos, A. Kanavos, Ph. Mylonas |
| A Hybrid Feature-Guided Attention Framework for Suicidal Ideation Detection |
| 6th International Conference on Novel & Intelligent Digital Systems (NIDS 2026), September 23-25, 2026, Athens, Greece |
|
ABSTRACT
|
| Accurate detection of suicidal ideation in social media content is important for supporting early risk identification in computational mental health. The widespread use of platforms such as Reddit and Twitter has enabled the analysis of large-scale, user-generated text, offering new opportunities to identify linguistic and emotional indicators of psychological distress. In this study, suicidal ideation detection is investigated through a hybrid feature-guided attention framework that combines contextual language representations with affective and linguistic features. The proposed model uses a BERT encoder to extract semantic information, an attention mechanism to highlight informative textual cues, and a gated fusion module to adaptively integrate auxiliary sentiment and language-use features. The experimental evaluation is conducted using publicly available Reddit and Twitter datasets, supporting both in-domain and cross-domain assessment. The model is evaluated using accuracy, precision, recall, and F1-score. The proposed framework aims to improve robustness in noisy and informal social media text. It contributes to computational mental health research by exploring hybrid neural architectures for risk-related text classification.
|
| 23 September, 2026 |
| G. Vonitsanos, A. Kanavos, Ph. Mylonas, "A Hybrid Feature-Guided Attention Framework for Suicidal Ideation Detection", 6th International Conference on Novel & Intelligent Digital Systems (NIDS 2026), September 23-25, 2026, Athens, Greece |
[
BibTex] [
Print] [
 Back] |
