Learning in Twitter Streams with 280 Character Tweets

Abstract

Social networks are a thriving source of information and applications are pervasive. Twitter has recently experienced a significant change in its essence with the doubling of the number of maximum allowed characters from 140 to 280. In this work we study the changes that come from such modification when learning systems are in place. Results on real datasets of both settings show that transferring models between both scenarios may need special treatment, as bigger tweets are harder to classify, making such dynamic environments even more challenging.