Depression detection for twitter users using sentiment analysis in English and Arabic tweets

Abstract

Abstract Depression, a common mental health issue, significantly disrupting an individual's daily functioning and increasing premature Depression, a common mental health issue, significantly disrupting an individual's daily functioning and increasing premature mortality risk. The ubiquitous use of social media platforms for expressing sentiments and sharing daily activities provides a fertile ground for early detection of depression. This paper makes significant contributions in utilizing online platforms for depression detection. Firstly, we introduce five machine-learning models to detect depression in Arabic and English text from Twitter. For Arabic text, our optimal model achieved a high accuracy with an F1-score of 96.6% for binary classification of depressed and non-depressed tweets. For English text, excluding negations, the model accomplished an F1-score of 92% for binary classification and 88% for multi-classification (depressed, indifferent, happy). When considering negations, the model demonstrated a slightly lower performance with an 87% and 85% F1-score for binary and multi-classification respectively. Secondly, we present three unique corpora: one manually annotated Arabic corpus, and two automatically annotated English corpora—with and without negation. These corpora encompass a broad spectrum of emotional sentiments, enhancing the depth of our analysis. Lastly, the paper presents a novel web application for depression detection, implementing our refined models. This application enables the identification of depression symptoms in tweets and prediction of an individual's depression trends, supporting both English and Arabic languages. This research represents a significant stride forward in mental health detection leveraging the widespread use of social media.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 53 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.