Representation Learning (RL) consist in automatically learn features from the raw data input. The main goal is to obtain useful representations for machine learning algorithms. The recent success of some works related to RL (e.g., DeepLearning) has attracted the interest of NLP scientist. Some of the most relevant works in NLP have to do with learning the representations of words, also called: Word embedding. Word embedding maps the words of the vocabulary to vectors of real numbers in a low dimensional space. In this talk, some of the most relevant works for word embedding are briefly presented. In specific, those works that motivated the successfully Word2Vec approach from Tomas Mikolov. More importantly, the key elements of Word2Vec are fully explained, which is very important to make it work properly, to improve it in new research or to adapt it for other areas in computer science. The talk is mainly based in Notes on Representation Learning made by López-Monroy during a 3-month research secondment at Universidad Politécnica de Valencia, Spain.