PAN23 Profiling Cryptocurrency Influencers with Few-shot Learning

This is the dataset for the shared task on Profiling Cryptocurrency Influencers with Few-shot Learning. Please consult the task's page for further details on the format, the dataset's creation, and links to baselines and utility code.

Task: In this shared task we aim to profile cryptocurrency influencers in social media, from a low-resource perspective. Moreover, we propose to categorize other related aspects of the influencers, also using a low-resource setting. Specifically, we focus on English Twitter posts for three different sub-tasks:

Low-resource influencer profiling (subtask1):

Input:
32 users per label with a maximum of 10 English tweets each.
Classes: (1) null, (2) nano, (3) micro, (4) macro, (5) mega

Official evaluation metric: Macro F1

Submission: TIRA.

Baselines: User-character Logistic Regression; t5-large (bi-encoders) - zero shot [7], t5-large (label tuning) - few shot [7]

Low-resource influencer interest identification (subtask2):

Input:
64 users per label with 1 English tweet each.
Classes: (1) technical information, (2) price update, (3) trading matters, (4) gaming, (5) other

Official evaluation metric: Macro F1

Submission: TIRA.

Baselines: User-character Logistic Regression; t5-large (bi-encoders) - zero shot [7], t5-large (label tuning) - few shot [7]

Low-resource influencer intent identification (subtask3):

Input:
64 users per label with 1 English tweets each.
Classes: (1) subjective opinion, (2) financial information, (3) advertising, (4) announcement

Official evaluation metric: Macro F1

Submission: TIRA.

Baselines: User-character Logistic Regression; t5-large (bi-encoders) - zero shot [7], t5-large (label tuning) - few shot [7]

Versioning: 

1.0: initial upload

1.1 fixed a minor bug where some users contained some non-English text. Since English is the target language in the competition, all non-English texts have been replaced or removed.