A BERT BiLSTM and BiTCN model for suicidal ideation identification in Spanish texts

Abstract

The timely detection of suicidal ideation is essential for reducing suicide-related deaths. Early signals of suicidal thoughts can often be identified through the analysis of textual content shared on social media platforms. In this context, Artificial Intelligence (AI) techniques, particularly Natural Language Processing (NLP), have shown strong potential for identifying linguistic patterns that enable the accurate classification of suicide-related content. However, NLP models developed for one language cannot be directly transferred to other languages due to their specific linguistic characteristics, highlighting the need for language-specific approaches. This study proposes a computational model for detecting suicidal ideation in Spanish-language social media posts. The proposed architecture combines Transformer-based representations with BiLSTM and BiTCN neural network layers to capture contextual and sequential linguistic patterns. The model is trained and evaluated using two corpora of different sizes composed of suicide-related posts collected from Twitter (X) and Reddit. The proposed model achieves an F1-score of 93.51% on the smaller corpus and 90.34% on the larger corpus, obtaining higher scores than the evaluated baselines and reimplemented comparison models under the same experimental setting. These results suggest that the proposed architecture is effective for suicidal ideation detection in Spanish-language social media texts within the two evaluated corpora. The consistent performance obtained on both datasets suggests that the model is capable of capturing relevant linguistic patterns associated with suicidal ideation and highlights its potential to support research and early screening initiatives in Spanish-speaking online environments.