Profiling Cryptocurrency Influencers using Few-shot Learning

Abstract

This research provides a novel method for identifying cryptocurrency influencers on social media in a low-resource environment. The analysis focuses on English-language Twitter messages and divides influencers into impact categories ranging from minimal to massive. With a maximum of 10 English tweets per user, the dataset consists of 32 people per category. By comparing the suggested approach to two baseline models—Usercharacter Logistic Regression and t5-large (bi-encoders) using zero-shot and label tuning few-shot methods—the proposed system is evaluated using the Macro F1 measure. The findings show that the suggested approach operates effectively in low-resource environments and has the potential to be used to further in-depth studies of influencer profiling. © 2023 Copyright for this paper by its authors.