README.md

---
license: mit
task_categories:
- text-classification
language:
- az
size_categories:
- 100K<n<1M
---


This dataset contains 150K (train + test) cleaned tweets in Azerbaijani. Tweets were collected in 2021, and filtered and cleaned by following these steps:

- Initial data were collected by using twint library. The tool is currently deprecated, cannot be used with new Twitter.
- On top of the already filtered data, I applied an additional filter to select Azerbaijani tweets with using fastText language identification model.
- Tweets were classified into 3 emotion categories: {positive: 1, negative: -1, neutral: 0} by using emojis as rule-based classifier.
- Tags, usernames, and emojis were later cleaned.
- Short tweets were filtered out.