INF2170H Chap.6 Search, Metadata and Retrieval
Search, Metadata and Retrieval
Trace misses to structural causes
Metadata describe objects and relationships so systems can organise, display and retrieve them. This chapter distinguishes descriptive, structural and administrative roles, then checks a failed lookup in order: object, field, value, vocabulary and retrieval route.
The goal is a reproducible repair rather than hidden keyword accumulation.
Maintain the language layer
Controlled terms support consistent filtering and aggregation, while synonyms preserve entry through the words people use. Scope, equivalence and related term decisions require ownership and review.
Search and browse share authoritative records but expose their descriptive choices through different interactions.
Model fields around object classes
Global fields should carry the same meaning across every object. Type specific fields belong only where they describe a relevant property. Requiring duration for a policy creates meaningless blanks, while omitting it from media weakens filtering.
Define field purpose, applicability, value type, repeatability, requirement and vocabulary. Separate absent, unknown and not applicable so quality review can distinguish missing work from a field that does not belong.
Test the schema with ordinary boundary records and a plausible future type. Ideal examples can hide ambiguous scope.
If contributors need to memorise exceptions, improve the field rule, validation or object model. Keep source values when normalising legacy metadata so a reviewer can trace how the controlled form was derived and reverse an incorrect mapping.
Repair retrieval through a preserved query
Record the exact words, filters, expected object and actual result set.
Confirm the object exists and is visible before changing vocabulary. Then inspect the applicable field, value form, synonym relationship and route configuration. A zero result can come from an unpublished record, missing metadata, inconsistent language or an index that ignores the field. Each cause calls for a different repair.
Review false positives along with misses.
Automatic substitution between merely related terms can increase recall while burying the useful item. Use equivalence only for genuine synonyms and consider suggestions or scoped expansion for related concepts. After a change, rerun the same query and a set of neighbouring queries. A fix is durable when it improves the intended need without corrupting browse facets, other object types or the maintained vocabulary.
What this chapter covers
- 01
Metadata roles
- 02
Object classes
- 03
Field rules
- 04
Controlled vocabulary
- 05
Synonym mapping
- 06
Retrieval testing
Worked example · free
Repair a missing result
- 2Confirm the object and visibility.
- 3Inspect fields and controlled values.
- 2Repair and rerun the query.
Key terms
- Descriptive Metadata
- Descriptive Metadata records attributes that support identification and discovery.
- Structural Metadata
- Structural Metadata records hierarchy and relationships among parts or objects.
- Controlled Vocabulary
- Controlled Vocabulary maintains preferred terms, relationships and scope for consistent values.
- Regression Query
- Regression Query is a preserved information need rerun after a retrieval change.
Search, Metadata and Retrieval FAQ
How should a zero result be diagnosed?
Confirm that the expected object exists, is visible and belongs in the corpus. Inspect whether the relevant concept is represented in an applicable field, whether the value follows the controlled vocabulary and whether the route indexes or filters that field.
When should a controlled vocabulary be used?
Use controlled values when consistent aggregation, filtering or cross collection retrieval matters. Define scope notes, relationships and ownership. Keep free text where nuance is essential, and map genuine synonyms so users do not need to know the preferred term before searching.
Why are related terms not always synonyms?
Synonyms represent equivalent concepts, while related terms merely have a meaningful association. Automatic substitution between related concepts can expand results unpredictably. Model the relationship and decide whether suggestion, scoped expansion or a browse link better supports the task.
Assessment move
Take one known item and one failed query. Trace object, field, value, vocabulary and route in sequence, then rerun the same query after the smallest repair. Compare search with faceted browse and keep shared metadata rules consistent. Create a compact record for the object and label every field as free text, controlled, identifier, relationship or administrative data.
State the information need and what would count as a relevant result before looking at the ranking. If the object is absent, inspect ingestion, field population, controlled terms, synonyms and indexing in order. If it appears too low, inspect which fields matched and whether irrelevant objects share overbroad language. Change one structural cause at a time, preserve the original query and test nearby queries for regressions.
Practise distinguishing a missing value from a missing vocabulary relationship and a missing object relationship, since each requires a different repair. Use facets for visible controlled dimensions and search for less predictable expressions, while keeping their underlying values aligned. Record who owns definitions, term changes and quality checks.
Finish by naming the exact failed route, the smallest evidence-based change and the result that would confirm or refute the repair.
