A library can own a million useful items and still fail if nobody can work out where anything belongs.
Classification solves that problem by giving the collection a shape. It turns an inventory into a map.
This article is part of eduKateSG’s How X Works programme and the How a Library Works series. The aim is not to memorise classification codes. It is to understand the mechanism that makes organised browsing possible.
The shortest useful answer
Library classification works by assigning material to structured subject positions so that related things become predictably adjacent and different things become meaningfully separated.
That sentence contains the whole machine.
- Assigning means a decision must be made.
- Structured means the positions follow a system rather than personal taste.
- Subject positions mean the system represents what an item is mainly about.
- Adjacency makes browsing useful.
- Predictability allows many people to use the same logic.
- Separation prevents the collection from collapsing into an undifferentiated mass.
Classification is not the same as cataloguing
The two processes are closely connected, so they are easy to confuse.
Cataloguing describes an item: title, creator, edition, publisher, date, identifiers, notes, physical or digital form, subjects and relationships. Classification gives the item a position inside an organised subject structure.
If cataloguing answers “what is this?”, classification answers “where does this sit?”
One produces identity. The other produces neighbourhood.
Why the shelf needs an intellectual architecture
Imagine shelving books simply in the order they were purchased. The newest acquisition goes at the end. Nothing is technically lost, but the shelf tells the user almost nothing.
Now imagine grouping materials so that mathematics sits with mathematics, geometry near other geometry, number theory near other number theory, and specialised subfields further divided within those groups. The shelf becomes informative before a book is even opened.
Classification converts physical proximity into semantic proximity.
That is why a classification number is more than a storage label. It is a compressed statement about an item’s place in a conceptual landscape.
The three jobs of a classification system
A useful library classification system normally performs at least three jobs at once.
1. It groups related material
Users looking for one topic can browse neighbouring items without knowing their titles or authors in advance. This is especially powerful when the user’s question is broader than a single known item.
2. It separates unlike material
A book about the chemistry of batteries should not normally sit beside a political history of elections merely because both titles begin with the letter B. Subject structure prevents accidental proximity from masquerading as meaningful proximity.
3. It generates a stable locator
Once a class position is converted into a call number and combined with additional information such as author or title coding, a specific item can be placed in a predictable sequence.
Hierarchy: moving from broad to narrow
Classification systems often work hierarchically. A broad domain divides into narrower domains, which divide again.
Consider a simplified path:
- Science
- Physics
- Mechanics
- Fluid mechanics
- Boundary-layer flow
Each step reduces the set. The user can stop at any useful level. This makes classification a narrowing mechanism.
Hierarchies also create inheritance. If something is classified under fluid mechanics, it is implicitly part of mechanics, physics and science. A well-designed hierarchy lets one specific placement carry several levels of meaning.
Dewey Decimal Classification: the logic of decimal expansion
The Dewey Decimal Classification is familiar in many public and school libraries because it divides knowledge into large numerical classes and then expands those classes with increasing precision.
The important mechanism is not the memorisation of every number. It is the idea that a notation can encode hierarchical position. A short number can represent a broad subject. Additional digits can represent narrower distinctions.
This produces a sortable sequence. The notation acts simultaneously as subject code and shelf-order instruction.
The elegance is operational: a large body of knowledge can be distributed across a numeric space that supports both expansion and browsing.
Library of Congress Classification: a different map for a different scale
Large academic and research libraries often use Library of Congress Classification. It combines letters and numbers and provides extensive subject divisions suited to large, specialised collections.
The deeper lesson is that there is no single inevitable way to divide knowledge. Classification systems are engineered responses to collection scale, institutional history, disciplinary structure and user need.
A primary school library and a national research library should not be expected to require identical granularity. A system can be logically sound and still be poorly matched to a particular collection.
A classification scheme is a model of knowledge, not knowledge itself
This distinction matters.
The world does not naturally arrive divided into library classes. Humans create those classes in order to make the world manageable.
Disciplines overlap. New fields emerge. Old terminology becomes unacceptable. Political borders change. Scientific understanding moves. Interdisciplinary subjects resist a single home. A book can be equally relevant to economics, sociology, psychology and public policy but still require one principal shelf position.
Classification therefore always compresses complexity. It is useful because it simplifies, and imperfect for the same reason.
The problem of the book that belongs in several places
Physical books expose a hard constraint: one copy can normally occupy only one shelf position at a time.
A book about the mathematics of music might belong under mathematics, acoustics, music theory or education. A history of vaccination might belong under medicine, public health, history or social policy.
Classification chooses a primary route. Cataloguing can then restore some of the lost dimensionality through multiple subject headings and access points.
This is a recurring principle in information systems: storage may require one location while discovery can support many routes.
Call numbers: turning class into exact shelf position
A classification number is often only part of the call number. Libraries add elements that distinguish one item from another within the same subject class.
The resulting call number can encode several things at once: subject position, author or title sequence, edition, date, volume, copy or special location.
The mechanical purpose is exactness. Classification gets the user into the right neighbourhood. The rest of the call number gets the user to the right address.
Why shelf reading matters
A classification system can be perfect on paper and still fail in practice if items are shelved incorrectly.
This is why libraries perform shelf reading: checking whether materials actually appear in the required sequence.
A mis-shelved book is an interesting kind of information failure. The catalogue may say it exists. The classification may be correct. The call number may be printed accurately. Yet the item becomes locally invisible because the physical state no longer matches the informational state.
Libraries therefore depend on alignment between record and reality.
Controlled vocabularies: classification beyond the shelf
Classification is closely related to controlled vocabulary.
Natural language is messy. People use synonyms, abbreviations, slang, old terminology and different spellings. A controlled vocabulary chooses authorised terms and records relationships among them.
This can connect a user’s search term to the library’s preferred term, guide users from broad topics to narrower ones, and distinguish words that look alike but mean different things.
For example, a catalogue may need to know that an everyday phrase and a technical term point toward the same concept. It may also need to distinguish a person’s name from a place with the same name.
Faceted classification: one item, several dimensions
Digital systems can escape some of the one-shelf constraint by allowing users to filter the same item along several dimensions.
An academic article may be simultaneously discoverable by subject, date, author, language, document type, institution, region and access status. Each facet becomes a temporary classification axis.
This is powerful because different users begin with different questions. One wants all recent articles. Another wants only Singapore material. Another wants systematic reviews. Another wants open-access material.
Facets let the library reorganise the visible collection around the user’s current constraint without permanently changing the underlying records.
Classification and search solve different problems
Search is often excellent when the user knows a phrase, title, author or keyword. Classification is powerful when the user knows the territory but not the exact object.
Search asks the collection to match a query. Classification lets the user move through an organised space.
The strongest library systems combine both. Search gets you to a likely node. Classification shows you the neighbourhood around it.
Classification changes as knowledge changes
No serious classification scheme can remain frozen forever.
New disciplines appear. Existing fields merge or split. Scientific concepts are revised. Social terminology changes. New technologies create subjects that older schedules did not anticipate. Libraries must add, revise and sometimes relocate classes.
This creates a difficult balance between stability and accuracy.
Change too slowly, and the scheme stops reflecting the world. Change too quickly, and users lose continuity while libraries face expensive reclassification work.
Classification is therefore maintained like infrastructure: conservative enough to remain usable, adaptive enough to remain meaningful.
Bias enters through the map
Because classification is a human model, it can inherit the assumptions of the people and institutions that created it.
Older schemes may centre one geography, language, religion, social order or disciplinary worldview. Subjects can be named in ways later recognised as inaccurate or harmful. Entire regions of knowledge can receive little space because they were marginal to the original institution.
Correcting this is not simply a matter of etiquette. Terminology affects retrieval. Hierarchy affects visibility. Placement affects which relationships users encounter.
The map influences what becomes easy to see.
Classification in digital libraries
Digital libraries still classify even when users never see a call number.
Classification may appear as database categories, repository collections, subject taxonomies, knowledge graphs, entity types, metadata schemas, recommendation clusters or machine-generated topic labels.
The essential problem remains unchanged: how do we create enough structure for meaningful navigation without forcing a complex world into categories that are too crude?
Classification for AI systems
AI makes classification simultaneously easier and more important.
Machine learning can suggest subjects, extract entities, cluster documents and detect relationships at scales that would be difficult to process manually. But automated classification can also reproduce training bias, confuse context, overgeneralise or create unstable categories that lack institutional meaning.
A robust library therefore treats machine classification as evidence rather than unquestionable truth. Human-governed vocabularies, provenance and review remain valuable because they provide a stable reference frame.
AI can help place material. The library still has to decide what the placement means.
A classification system is a compression engine
The deepest way to understand classification is as compression.
A full book may contain hundreds of pages, many topics and several arguments. A classification code reduces that complexity to a compact statement about primary position. A subject heading reduces another part of the work to a standard phrase. A facet reduces a dimension to a filterable value.
Compression is necessary because users cannot read every item before deciding whether it is relevant. Classification creates a low-cost preview of intellectual location.
Good compression preserves the distinctions that matter. Bad compression hides them.
The complete mechanism
- The library defines or adopts a classification scheme.
- The scheme divides knowledge into organised subject regions.
- A cataloguer analyses what an item is mainly about.
- The item is assigned to the most useful class position.
- The class notation is combined with additional elements to produce a call number or locator.
- The item is placed in a sequence with related material.
- Subject headings and other metadata provide additional routes when one shelf position is insufficient.
- Search, browse and facets expose the structure to users.
- Errors, new subjects and changing language feed back into maintenance of the scheme.
That is how library classification works.
It is the discipline of deciding which differences matter enough to separate, which similarities matter enough to group, and how those decisions can be encoded so that another person can navigate the collection without first knowing what it contains.