Analyzing Textual Information at Scale

We provide an overview on the recent advances in textual analysis for social sciences. Count-based economic model, structured statistical tool, and plain-vanilla machine learning apparatus each have their own merits and limitations. To take a data-driven approach to capture complex linguistic structures while ensuring computational scalability and economic interpretability, a general framework for analyzing large-scale text-based data is needed. We discuss the recent attempts combining the strengths of neural network language models, such as word embedding, and generative statistical modeling, such as topic modeling. We also describe typical sources of texts and the applications of these methodologies to issues in finance and economics and discuss promising future directions.