Abstract:

Abstract Suicidal ideation is one of the most serious issues confronting today's youth, owing to mental illnesses such as anxiety, bipolar disorder, and depression. Suicidal ideation is difficult to categorise due to the way people use language and express themselves on Twitter or other social media platforms. Personal contextualization of such originality is difficult to accurately identify users at risk. Thus, intervening in the early stages of suicidal ideation can help to reduce the number of suicides. In this study, a Relational Graph Attention Network (RGAT) with a DNN classifier is used to detect suicidal ideation. Initially, Twitter posts are collected and preprocessed to remove unnecessary data. Bi-directional Encoder Representations from Transformers (BERT) are used for embedding the tweeter post to relate to the topic and sentimental words. The preprocessed data is then fed into Latent Dirichlet Allocation (LDA) and sentiment-based Lexicon approach to find the topic and extract sentimental words. RGAT relates the features from LDA and BERT as well as the lexicon-based sentiment approach. Both of these RGAT outcomes are concatenated and classified using the DNN algorithm. The performance metrics evaluate and compare the proposed method to existing models. The attained performance metrics like precision, accuracy, recall, and error for the proposed model are 88.22%, 88.21%, 88.19%, and 12%, respectively. The evaluated metric values of the proposed model is better than the values of the existing models. Thus the designed model using RGAT with a DNN classifier performs better and accurately detects suicidal ideation.

Keywords: