What Is Word Embedding and How Can It Be Useful for Social Science Research

• In Event: Embeddings

Abstract

Word embedding has gained significant interest and popularity among social scientists in the past five years. However, the application of word embedding to research is still sparse and limited. Most applications in political science use pre-trained word vectors as feature representations. In this paper, we demonstrate that word embedding can be useful for various research goals in addition to feature representation and semantics learning. Specifically, we show how to use word embedding for automated dictionary generation, multi-lingual text analysis, and for measuring complex concepts and biases in texts. In doing so, we offer a detailed explanation and tutorial of how to train word embedding models like word2vec, a neural network-based method. We also introduce some of the more recent embedding methods, including BERT and other transformer networks.

Authors

• Yaoyao Dai, University of North Carolina, Charlotte

• Benjamin J Radford, University of North Carolina at Charlotte