Lesson 2: Embeddings, where a sentence becomes a point

Module 4 · Lesson 2 · About 40 minutes · Needs: Lesson 1 of this module (what a network outputs) and NumPy from Module 1

The situation

Priya has a new worry. Module 3's classifier only knows the words it was trained on. A customer who writes "the hoover won't charge" gets no match for "vacuum" and lands in "other". Dev asks the same thing from the other side: which help article is closest to a ticket, with no rules?

Both problems have one answer. Turn every sentence into a list of numbers, an embedding, so that similar meanings get similar numbers. This lesson builds that tool, using one ticket all the way through: "The vacuum stopped charging after 3 months."