Urdu Depression Severity Dataset: 4,000 Trilingual Twitter Posts

4,000 Twitter posts in Urdu, English, and Roman Urdu are labeled for depression severity. The dataset features 4-class severity labels verified by both large language models and human annotators. It was created in 2024-2025 and shared on Kaggle.

Use Cases

• Train classifiers to detect depression severity based on multilingual social media text.
• Benchmark the performance of LLMs on mental health annotation tasks in low-resource languages.
• Analyze linguistic markers of depression across Urdu, English, and Roman Urdu text.
• Study cross-lingual transfer learning for mental health applications.

Strengths

• 4,000 data points provide a substantial corpus for model training.
• Labels are verified through a combined LLM and human annotation process.
• Covers three language scripts (Urdu, English, Roman Urdu), enabling multilingual analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal and platform bias inherent to Twitter data from Kaggle.

Provenance

Source

Kaggle

Collection Method

Collected from Twitter (X), with labels applied via LLM and human verification.

Time Range

2024-2025