Skip to main content
Search & Retrieval

Relevance Tuning

Relevance tuning is the systematic process of improving search ranking by adjusting signals, weights, and rules based on evidence and evaluation.

Also known as: Ranking tuning, Relevance optimization, Search tuning

Definition

Relevance tuning is the practice of improving how a search engine ranks results for real user queries. It includes adjusting field weights, boosting trusted sources, refining analyzers and synonyms, and validating changes with offline judgments and online analytics.

Why it matters

  • User trust: small ranking errors can break confidence in high-stakes domains.
  • Consistency: reduces “random-feeling” results across similar queries.
  • Business impact: better relevance improves conversions and reduces support load.
  • Safety: tuning can prefer authoritative sources and reduce risky matches.

How it works

Measure -> hypothesize -> adjust ranking -> evaluate -> deploy -> monitor

Good tuning starts with clear success metrics (time-to-result, satisfaction, task completion) and a representative query set.

Practical example

If users searching “WIB92 article 26” often click the official text, you can boost pages with official citations and demote generic commentary for that intent.

Common questions

Q: Should we tune with rules or machine learning?

A: Both can work. Rules are transparent and fast to iterate; ML can capture complex patterns but needs data and careful evaluation.

Q: What’s the fastest way to find tuning opportunities?

A: Look at search analytics for zero-result queries, high reformulation rates, and queries with low click satisfaction.

References

Manning, Raghavan & Schütze (2008), Introduction to Information Retrieval.