Depression Detection Using Social Media Analysis

Depression is a psychological disorder which impacts numerous people throughout the globe. In current society, individuals verbalize their emotions on social media, which can serve as a great tool to detect signs of depression through sentiment analysis. A deep learning system designed in this study aims to determine depression in people. The main focus is on preprocessing and feature extraction from a dataset of text messages labeled as “depressed” or “not depressed.” For preprocessing, the text was cleaned, broken into smaller parts (tokenization), removed unnecessary words (stop words), applied lemmatization, and did Part-of-Speech (POS) tagging using NLTK and SpaCy. Things like word usage, sentence length, and how often different types of words appear to understand the language patterns of depressed and not depressed individuals taken into consideration. To make the text data more structured, the TF-IDF transformation was used. To understand the semantic meaning of words in a way that deep learning models, processing of word embeddings with Word2Vec Model was also done. The obtained features moved from the extraction process into both LSTM and GRU computational frameworks. The LSTM produced accuracy results at $\mathbf{0. 9 0 6}$ along with the GRU model reaching 0.9134 accuracy levels. The successful outcome of our method shows how it can effectively identify depression cases in textual information.