Module 3 · Lesson 3 of 7
Mixing word search and meaning search
Word search found 0, meaning search 3, and the merged list starts with: I was charged twice for the same order this month.
Word search
No ticket shares a word with this question.
Meaning search
- RefundsI was charged twice for the same order this month.
- RefundsMy package arrived damaged and I need a refund for it.
- RefundsThe item does not match the listing, so I'd like a reimbursement.
Hybrid: both, merged
- RefundsI was charged twice for the same order this month.meaning #1 = 0.016
- RefundsMy package arrived damaged and I need a refund for it.meaning #2 = 0.016
- RefundsThe item does not match the listing, so I'd like a reimbursement.meaning #3 = 0.016
What this simplifies. The tickets and their places on the map are the same as in the meaning map, placed by hand. Real systems use embeddings computed by a model, and a keyword index such as BM25, which also weighs how rare each word is. Each result shows its topic, so you can spot a wrong match, such as a ticket that only shares the word stuck. Each search passes its top 5 to the merge, and the merged score is 1 ÷ (60 + rank) from each list.
Search by meaning finds a refund ticket for the question "money back", even with no word in common. But it can miss exact terms, such as an error code or a product name.
Word search, also called keyword search, is the opposite: it finds exact terms, but misses different words with the same meaning.
(Running a word search and a meaning search together, then merging their results, so a passage found by either can reach the top.) runs both and merges the two ranked lists, so a (A short passage cut from a longer document, so it can be turned into an embedding and found on its own.) found by either one can reach the top.
A common way to merge is reciprocal rank fusion: each list gives a chunk 1 ÷ (60 + its rank), and the scores add up. The 60 is a common setting, not a fixed rule.
In system design: Many (The live system that real users reach, as opposed to a test or practice copy.) systems use hybrid search, because people mix exact names with everyday words.