Skip to content
Text as Data and NLP for Finance / Embeddings and classification

Text Classification

A lexicon is a mapping the researcher writes down. Supervised classification learns the mapping from documents that humans have already labelled. For a 10-K or call tone feature, the measurement decision is the set of categories in the codebook. The implementation is the representation plus a classifier. The validation deliverable is accuracy on labelled documents that were not used to fit the model, with the training labels themselves released so the result can be rebuilt.
Learning objectives. By the end of this note you should be able to:
  • separate keyword counting from supervised classification;
  • write the naive Bayes ranking under a conditional-independence assumption;
  • state why a finance naive Bayes tone model is hard to replicate when the training documents and filters are unpublished;
  • name the held-out comparison that must exist before the label is used as an event feature.

In this lesson

  1. Two maps from text to a label
  2. Naive Bayes
  3. What finance applications have actually published
  4. The validation deliverable
  5. Coding the training set

Before you start

The full lesson, with worked examples and exercise solutions, is for members.

Read the full lesson

New to SelfMadeQuant.com? See membership options or browse the Text as Data and NLP for Finance syllabus.