Hybrid search for names, codes and meaning

Learning outcome: Combine retrieval signals when exact matches matter.

Lexical retrieval matches words and terms; semantic retrieval uses representations of meaning. Hybrid search combines signals from both. It can be useful when a query includes an exact identifier alongside a loosely worded description.

Do not assume combining methods always improves your task. Create questions that test exact codes, paraphrases and ambiguous terms. Compare the retrieved candidates before inspecting the final answer. Ranking scores from different systems may have different scales, so a naive average can be misleading. Use the retrieval system's supported combination method and evaluate the resulting ordering.

Worked example

A fictional user asks, “How do I reset the ZX-17 after the amber light stays on?” Semantic retrieval finds a useful reset description for ZX-71, while lexical retrieval finds ZX-17 but ranks a spare-parts page first. A hybrid candidate set may help, but the final evidence still needs the correct model and symptom.

Try it yourself

Create four queries about two similar device codes, including one paraphrase of a symptom. Define the correct source for each before looking at search results.

You can do this exercise on paper. If using an assistant, keep to the fictional material and an approved tool.

Check your understanding

What must be preserved when a query names ZX-17?

1. Only the general topic of resetting devices
2. Whichever model has the clearest instructions
3. The exact identifier as well as the intended task

Workbook

Download the course workbook to record your answer and compare it after the knowledge check.

Further reading

• Elastic: hybrid search
• Elastic: semantic reranking