Perceiving the level of depression from web text

Depression is one of the deadliest diseases found in today's world, and unfortunately, it is also one of the most ignored problems. Depression is a fact that is very hard to accept for any individual and is always a multistep process. The initial stage of Depression is Loneliness, and thus the information about these emotions can be leveraged and can help in the early detection of Depression, which in turn leads to suicidal thoughts. Tweet data analysis is one of the most popular ways to determine the presence of depression and suicidal thoughts, through the concepts of Machine Learning. Twitter proves to be a very rich source of data, as their user base is potentially large enough, but is also increasing in a fast manner. For the scope of this paper, we predicted from a user's specific tweet, which is categorized for loneliness. These tweets are analyzed to check the level of depression as moderate or severe when people start thinking of suicide. The simulation is carried out using four different models for one level of classification and eight models are used at the second level of classification. It is observed that Gated Recurrent Unit with BERT outperformed all the models and showed the accuracy of 99% and 97%. However, for class-1 recall with XLNet gave the best result with class-1 recall being 0.99. This application can help the individual in early detection of depression without any human intervention and seek medical help. Moreover, it also provides an insight about the feelings of the individual to the medical practitioners, which, in turn, can help them provide better decision-making.