NLP FOR DEPRESSION DETECTION BASED ON SOCIAL MEDIA POSTS: A SYSTEMATIC REVIEW

Abstract

ABSTRACT
This systematic review explores recent developments in natural language processing (NLP) for
depression detection using social media posts. Depression is a widespread mental health disorder
with challenges in early diagnosis due to stigma and limited clinical resources. Social media
platforms offer rich textual data reflecting users’ mental states, enabling computational methods for
timely detection. This review synthesizes 41 peer-reviewed studies from 2020 to 2025, focusing on
datasets, preprocessing, feature extraction, machine learning and deep learning models, evaluation
metrics, explainability, and ethical concerns. Transformer-based models such as BERT and BiLSTM
show superior performance by capturing contextual information. Key linguistic indicators include
increased use of first-person pronouns and negative emotion words. Despite promising results,
challenges remain in handling informal language, data privacy, and annotation quality. Explainable
AI techniques are increasingly adopted to improve transparency and clinical trust. Future research
should prioritize multilingual datasets, privacy-preserving approaches, and multimodal data
integration to enhance detection accuracy and applicability. Overall, NLP-based depression detection
from social media holds significant promise as a scalable, cost-effective tool for early mental health
intervention.