Coresets in NLP

Abstract

Coresets are small, weighted summaries of large datasets such that solutions found on this summary are provably competitive with those found on the entire dataset. This work studies the use of coresets in the field of Natural Language Processing. We observe how the theoretcial concept of coresets works in practical use cases by applying it in the NLP problems of Emotion Classification and News Article Classification as we try to merge large collections of unlabelled and mostly garbage data with meaningful datasets to create more well-informed and diverse models.