Emoji are ubiquitous in digital communication: over 1800 different pictographs can decorate our messages, helping convey intent or mood better, or at least more succinctly, than plain words. How can we recommend the most relevant emoji to a user who is composing text? We train several machine learning models to predict emoji, leveraging a corpus of public Twitter data and using word embeddings and sequence-based neural networks to capture a rich vocabulary and phrase-level semantics. I will present our approach, insights about working with emoji, and strategies for optimizing for mobile and continual performance improvement by training on user feedback.
Stacey Svetlichnaya is a software engineer on the Yahoo Vision & Machine Learning team. Her recent deep learning research includes object recognition, image aesthetic quality and style classification, photo caption generation, and modeling emoji usage. She has worked extensively on Flickr image search and data pipelines, as well as automating content discovery and recommendation. Prior to Flickr, she helped develop a visual similarity search engine with LookFlow, which Yahoo acquired in 2013. Stacey holds a BS and MS in Symbolic Systems from Stanford University.