Scholarlink

Abstract <p>Suicidal ideation (SI) represents a serious public health issue that demands early interventions to prevent devastating consequences. This study proposes a specialized word embedding model designed for the automatic detection of SI in texts, integrating advanced natural language processing (NLP) techniques, optimization through bioinspired algorithms, and incremental embedding training. The model was trained exclusively on phrases related to SI, enabling it to capture complex emotional and contextual patterns associated with this risk. For classification, a neural network architecture was employed that combines convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM), achieving an accuracy of 97.9%, a specificity of 98.5%, and a sensitivity exceeding 97%. These results highlight the model’s capacity to reduce false positives and effectively detect high-risk cases. This study demonstrates that the integration of specialized word embeddings with bioinspired optimization techniques and incremental embedding training constitutes a robust and effective strategy for the early detection of SI. The findings highlight the potential of this approach for implementation in real-time monitoring and prevention systems, contributing significantly to mental health efforts.</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

Specialized Word Embedding Model for the Detection of Suicidal Ideation Using Deep Learning

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Lizethe-Guadalupe Reyna-Morán,
  • Francisco-Javier Luna-Rosas,
  • Julio-Cesar Martínez-Romo,
  • Iván Castillo-Zuñiga,
  • Marco-Antonio Hernández-Vargas,
  • Mario-Alberto Rodríguez-Díaz

摘要

Abstract

Suicidal ideation (SI) represents a serious public health issue that demands early interventions to prevent devastating consequences. This study proposes a specialized word embedding model designed for the automatic detection of SI in texts, integrating advanced natural language processing (NLP) techniques, optimization through bioinspired algorithms, and incremental embedding training. The model was trained exclusively on phrases related to SI, enabling it to capture complex emotional and contextual patterns associated with this risk. For classification, a neural network architecture was employed that combines convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM), achieving an accuracy of 97.9%, a specificity of 98.5%, and a sensitivity exceeding 97%. These results highlight the model’s capacity to reduce false positives and effectively detect high-risk cases. This study demonstrates that the integration of specialized word embeddings with bioinspired optimization techniques and incremental embedding training constitutes a robust and effective strategy for the early detection of SI. The findings highlight the potential of this approach for implementation in real-time monitoring and prevention systems, contributing significantly to mental health efforts.