Stress and Depression Classification in Social Media using Contextual Knowledge Attention based Gated Recurrent Network.

The detection of stress and depression on social media focuses on analyzing social media posts to identify signs of stress and tendencies toward depression. It aims to offer early detection of the mental health states. However, detecting signs of stress or depression through social media posts presents significant challenges due to the unstructured nature of the text in the posts, and the diverse language styles used by individuals. So, this research proposes Contextual Knowledge Attention mechanism-based Gated Recurrent Unit (CKA-GRU) for the classification of posts from a popular social media blog site. The CKA mechanism dynamically applies weights to various parts of the text, allowing the GRU to prioritise important words based on stress and depression, which enhances classification accuracy. The effectiveness of the proposed CKA-GRU approach is evaluated on two standard datasets: Dreaddit and Depression_Mixed. These datasets consist of user posts analyzed to detect patterns indicative of stress or depression on social media. The relevant features are extracted from the pre-processed data using the Bag of Words (BoW). Finally, the CKA-GRU approach is employed for the classification of posts into binary classes for enhancing the overall performance of the model. The experimental results demonstrate that the proposed CKA-GRU method attains a commendable accuracy of 91.39% and 95.20% on the Dreaddit and Depression_Mixed datasets, respectively. These results prove that the proposed CKA-GRU approach accomplishes superior outcomes than the existing approaches namely, Bow with Logistic Regression (BoW-LR) and Knowledge-aware and Contrastive Network (KC-Net).