Deep Learning based Detection and Prediction of Suicide Ideation using Multimedia Data Fusion

Social media platforms have become crucial outlets for individuals to express their thoughts, including those related to suicidal ideation. Detecting such expressions early is vital for preventing potential suicide attempts. This research aims to demonstrate the feasibility of automatically identifying messages containing suicidal thoughts using machine learning and deep learning classification algorithms applied to social media data, specifically from Reddit. To address the challenge of early detection, we utilized a combination of Long Short- Term Memory (LSTM) and Convolutional Neural Networks (CNN). In this paper, LSTM–CNN model was employed to analyze and evaluate the performance of various classification models. Reddit posts were collected and preprocessed for analysis, with the combinatorial neural network architecture utilizing word-embedding technology to enhance relevance ranking and classification accuracy. The results indicate that the LSTM–CNN combination model significantly outperforms traditional methods in detecting suicidal ideation. The deep learning architecture proved to be both reliable and effective in developing an efficient model for identifying the likelihood of suicide in text data. These findings highlight the potential of deep learning models, particularly those integrating LSTM and CNN with word embedding, to significantly improve the detection and prediction of suicidal ideation on social media platforms, offering promise for early intervention and mental health care efforts.