Journal of Computational Social Science, Springer Nature · Published
Detecting Depression: Employing Word-Embeddings and Sentence Transformers
Built depression-detection models using word embeddings (GloVe) and sentence transformers (T5, MiniLM, MPNet), raising the best F1-score from 93.24% to 98.04% on a curated dataset.
Feature extraction and classification were optimized via Bayesian Optimization (Optuna) and multi-classifier ensembles. The paper also discusses the ethics of ML-based depression detection, AI-human collaboration, and psychiatric applications.