CourseNatural Language Processing: The Complete Course · Module 6: Word and Sentence Embeddings · part 43 of 73
Part 43 · Module 6: Word and Sentence Embeddings

Lesson 4: Semantic search: finding the right help article

43 min read·14 Oct 2026
Module 6 · Lesson 4 · About 50 minutes · Needs: basic Python; TF-IDF and BM25 (Module 4 Lesson 2) and cosine similarity help; this lesson loads its own encoder

The situation

It is Monday 12 October 2026, project week 6. A customer asks Ines whether she can spread the cost of a Volt 2. Ines knows the help centre covers it, types "pay in instalments" into the keyword search, and gets the EU delivery article first. The right one, KB18 "Paying for your order", comes 20th of 24. It says "spread the cost over 12 months" and never uses the word "instalments".

Rosa's fourth job is to suggest the right article while Ines replies. This lesson builds search by meaning, scores it against keyword search on 60 labelled queries, and finds where each one fails (simulated data).

KEYWORD SEARCH: "PAY IN INSTALMENTS" 1 KB09 EU delivery 2 KB01 Returning an item 3 KB04 Warranty ... 16 more articles 20 KB18 Paying for your order the answer says "spread the cost over 12 months" never "instalments"
Ines's query puts the EU delivery article first (red: wrong) and the article that answers it 20th (green: the right one), because the answer uses other words.

The rest of this part opens with an account

Signing up is free and takes one email. Nothing to pay, and nothing renews.