Lesson 2: Three ways to turn text into numbers you can search
19 min read·13 Oct 2026
Module 5 · Lesson 2 · About 45 minutes · Needs: Lesson 1 of this module (tokens) and Module 4 Lesson 2 (embeddings and cosine)
The situation
Priya has a request. The help-centre search box only finds articles containing the exact words a customer typed: "money back" finds nothing, although HC-02 answers it. She wants search that understands what people mean.
Dev wants it before the answer engine (Module 7), which will use the same search to pick articles (Module 8). This lesson builds it over the 12 help articles three ways: weighted word counts (TF-IDF), the same counts with two corrections (BM25), and learned vectors (embeddings). Three real tickets are the queries.