Hands-On Machine Learning for Algorithmic Trading

Labeling

Many NLP applications learn to predict outcomes from meaningful information extracted from text. Supervised learning requires labels to teach the algorithm the true input-output relationship. With text data, establishing this relationship may be challenging and may require explicit data modeling and collection.

Data modeling decisions include how to quantify sentiments implicit in a text document like an email, a transcribed interview, or a tweet, or which aspects of a research document or news report to assign to a specific outcome.