Abstract
Detecting the emotional content of text is one of the most popular NLP tasks. In this paper, we propose a new methodology based on identifying "idealised" words that capture the essence of an emotion; we define these words as having the minimal distance (using some metric function) between a word and the text that contains the relevant emotion (e.g. a tweet, a sentence). We look for these words through searching the space of word embeddings using CMA-ES. Our method produces state of the art results, surpassing classic supervised learning methods.