Deep Learning Approaches for Early Detection of Depression using Sarcasm

Abstract

The huge popularity of social media like Facebook, Twitter, WhatsApp and Instagram where most people can share their opinions about other people without any social dishonours. This can lead to people being depressed. Furthermore, sarcastic words have a significant impact on depression levels. As a result, early depression detection is critical. Despite the fact that depression detection was done by using algorithms such as SVC, NB, DT, and LR have been developed using the Twitter dataset and sarcastic statements. However, there is scope for improvement. The proposed model for effectively detecting sarcasm is referred to as " Sarcastic News Dataset and Tweet-based Depression Detection (SNTDD)”. The proposed model for detecting sarcastic remarks in text data uses ensemble Deep learning models and compared it with the machine learning models. The proposed model is used the ensemble model of Deep learning model and the dense layer, the model gives better accuracy and reduce the loss. Additionally, it works on positive sarcastic text to increase performance. The significance of this is that more positive indicates it has a stronger impact on mental health or raises the amount of depression. Sarcasm in study can be difficult to identify as a result of the complex relationship of both positive and negative characteristics. In contrast to signs such as tone or facial expressions, natural language processing addresses a challenge of detecting sarcasm with requiring the use of context markers. In contrast to signs such as tone or facial expressions, natural language processing addresses the difficulty of detecting sarcasm without the need of context markers. The experimental results reveal that the suggested model "Sarcastic News Dataset and Tweet-based Depression Detection (SNTDD)" is tested on the data. The model outperforms deep learning and machine learning algorithms on the news headline dataset, with an accuracy of 97.4%. Consequently, the proposed model received a 94.4% F1 score.