Research Area:  Machine Learning
Word embedding, where semantic and syntactic features are captured from unlabeled text data, is a basic procedure in Natural Language Processing (NLP). The extracted features thus could be organized in low dimensional space. Some representative word embedding approaches include Probability Language Model, Neural Networks Language Model, Sparse Coding, etc. The state-of-the-art methods like skip-gram negative samplings, noise-contrastive estimation, matrix factorization and hierarchical structure regularizer are applied correspondingly to resolve those models. Most of these literatures are working on the observed count and co-occurrence statistic to learn the word embedding. The increasing scale of data, the sparsity of data representation, word position, and training speed are the main challenges for designing word embedding algorithms. In this survey, we first introduce the motivation and background of word embedding. Next we will introduce the methods of text representation as preliminaries, as well as some existing word embedding approaches such as Neural Network Language Model and Sparse Coding Approach, along with their evaluation metrics. In the end, we summarize the applications of word embedding and discuss its future directions.
Understanding Natural Language
Author(s) Name:  Yang Li, Tao Yang
Journal name:  Guide to Big Data Applications
Publisher name:  Springer
Volume Information:  pp 83-104
Paper Link:   https://link.springer.com/chapter/10.1007/978-3-319-53817-4_4