Dataset origin: https://zenodo.org/records/4638948

Description

TwiSty is a corpus developed for research in author profiling. It contains personality (MBTI) and gender annotations for a total of 18,168 authors spanning six languages. We distribute the Twitter ids of these authors as well as the ids of their available tweets at the time of corpus development. The tweets have undergone language identification and can be found in a 'confirmed' (as belonging to the language in which the author is situated) and 'other' category (other tweets of this author, automatically identified to be in other languages).

The corpus consists of json files, one for each of the six languages: Dutch, Italian, German, French, Portuguese, Spanish.

Each json file contains Twitter profiles with their self-assessed MBTI personality type and annotated gender. Each profile has two lists of tweet ids: one of which the language was confirmed to belong to this corpus, the other contains all the other tweets we were able to mine.

Each json file is structured as follows:

{user_id1 : 	
        {`user_id': user_id1,
        `mbti': `ESTP',
        `gender':'M',
        `confirmed_tweet_ids': [tweet_id1, tweet_id2, tweet_id4],
        `other_tweet_ids': [tweet_id3, tweet_id5]
        }
}

Citation

Verhoeven, B., Daelemans, W., & Plank, B. (2016). TwiSty: A multilingual Twitter 
Stylometry Corpus for gender and personality profiling. In: Proceedings of the 10th 
International Conference on Language Resources and Evaluation (LREC 2016). Portorož, 
Slovenia. European Language Resources Association (ELRA). 1632-1637.