Depression Mimicking Expressions

 The proposed dataset is designed to address the growing use of social media platforms for expressing mental health struggles, including severe depression. Existing datasets often suffer from an imbalance between depressive and non-depressive instances and fail to account for depression-mimicking expressions—such as stress, anxiety, sadness, sarcasm, and complaints—which are frequently misclassified as severe depression due to linguistic similarities. This leads to a high rate of false alarms and undermines the reliability of detection systems. To overcome these limitations, this dataset includes newly curated and annotated social media posts that not only identify instances of severe depression but also distinguish them from depression-mimicking expressions. The curated dataset makes up 60% of the total, while the remaining 40% is sourced from existing datasets focused on severe depression (refer to the links below). Together, this combination yields a total of 38,017 text samples. This dataset aims to provide better training for machine learning models and facilitate more realistic evaluations for accurately detecting severe depression.MHB: https://www.dropbox.com/scl/fo/y7p04go4bvvcwojalcwie/AAhiHEVm8jj_mVcGJB7JmPc?rlkey=gvl1q8zgru1drkloa0og27fgq&e=1&dl=0Dreaddit: https://github.com/gillian850413/Insight_Stress_Analysis/tree/master/data GoEmotions: https://github.com/google-research/google-research/tree/master/goemotions/data