The Science of #Hashtags
Event Type
Featured Speaker
Broadening Participation in Computing
TimeWednesday, September 155:45pm - 6:15pm CDT
DescriptionA Tweet contains information that goes beyond 280 characters. Just the text in a Tweet provides meaningful information enriched with emojis, #hashtags, and @mentions. However, a Tweet also contains other modalities such as images, likes, replies, and trends, each of them representing a piece of a puzzle. Using Machine Learning to learn to combine these signals leads to a global and better understanding of Tweets. In this talk, I present different strategies for learning multi-modal representations from Tweets using Transformers and BERT neural architectures. I also describe some techniques needed for deploying these heavy models in production.