Suicidal ideation is regarded as one of the major mental health concerns that not only negatively effects one's life but also has an unfavorable impact on society. As a result, early suicidal ideation identification has long been viewed as an essential undertaking that may assist both society and suicidal individuals. In this sense, social media material has been demonstrated to suggest early indicators of this condition because it provides a channel for individuals to express their views and beliefs. Natural language processing approaches, in conjunction with deep/machine learning classification models, have proved effective in extracting behavioral and textual aspects from social media posts as a growing study topic. In this project, three different models based on GAT, GCN and Sage have been implemented to examine their performance on two prominent social networking platforms, namely Twitter and Reddit.
In the current trials, four datasets are utilized to evaluate the performance of the approaches, including textual material obtained from both prominent social networks Twitter and Reddit. To begin, Twitter is a well-known online social networking tool that was created in 2006 and receives hundreds of millions of tweets everyday. Similarly, Reddit is a social news forum comprised of multiple sub-communities each with their own specialized topic, where site members may exchange and upload their content to be voted on by other users. To protect users' privacy, the following datasets include no personal information since such information has been replaced with a unique ID or the text of the postings has been utilized anonymously. All datasets include binary labeled suicidal and non-suicidal tweets and Reddit posts.