Mapping Knowledge: Citation Analysis and Graphic Displays

Knowledge does not arrive in neat boxes. A single research paper on water pollution might draw on chemistry, public health, economics, and law all at once. So how do we actually see the shape of a subject – its branches, its overlaps, and the hidden bridges between fields that look unrelated on a shelf? For students of library and information science, this is the heart of mapping the structure of subjects: using the trail of citations that scholars leave behind, and turning that trail into a picture. This post explains how citation analysis reveals the architecture of knowledge, why old classification systems struggle to keep up, and how graphic displays make these invisible relationships visible.

Why we map the structure of subjects 🔗

Every discipline has an internal structure. Some topics sit at the core, some at the edges, and many connect to neighbouring fields. Traditionally, this structure was inferred by experts who read widely and formed a mental picture of how a subject was organised. The problem is that human judgement is slow, partial, and hard to verify. As scientific output exploded after the mid-twentieth century, no single scholar could read enough to map a field reliably.

This is where citation analysis changed everything. When researchers write papers, they cite earlier work. Those references are not random. They signal influence, agreement, debate, and intellectual lineage. If we collect millions of these references and study the patterns, we can reconstruct the structure of a subject from evidence rather than opinion. The foundation for this was laid when Eugene Garfield proposed citation indexes for science in 1955 and later built the Science Citation Index, creating the raw data that made large-scale knowledge mapping possible.

Citation analysis for structuring knowledge 🔗

Citation analysis treats the reference list as data. By measuring how documents, authors, and journals link to one another, it identifies clusters of related work and the gaps between them. The result is a structural view of a field: which ideas group together, which authors anchor a specialty, and how separate areas connect.

How citations reveal relationships 🔗

There are three main techniques, and each captures a different kind of relationship. The first is co-citation analysis. Two earlier documents are said to be co-cited when a later paper cites both of them together. The more often they appear together, the more closely related their ideas are assumed to be. Henry Small is regarded as the originator of this method, and when combined with clustering and multidimensional scaling, co-citation can map the structure of specialised research areas and even science as a whole.

The second technique is bibliographic coupling, which works in the opposite direction. Two papers are coupled when they cite the same earlier source. The more references they share, the more likely they tackle a similar problem. The third is co-word analysis, which looks not at citations but at the words and key phrases authors use. By tracking which terms appear together across many texts, this method maps the relationships between ideas within a subject area, offering a view of content rather than influence alone.

Identifying the research front and the intellectual base 🔗

Citation patterns also reveal where a field is moving. Scholars tend to focus their attention on a small, recent slice of the literature, which creates dense clusters of frequently cited work. These dense regions point to active “research fronts.” Co-citation analysis is essentially an attempt to locate these high-density areas within a citation network by grouping highly co-cited documents. This gives us two useful layers: the foundational works that a community keeps returning to, and the leading edge where new questions are being asked.

What makes this approach powerful is its scale and objectivity. Instead of one expert’s view, citation analysis aggregates the collective behaviour of thousands of researchers. Mapping a knowledge domain in this way draws on inputs such as network analysis, concept extraction, citation analysis, and visualization techniques, and the broad purpose of mapping a knowledge domain is to answer our questions about how a field is organised – useful to specialists and newcomers alike.

Challenges in traditional classification 🔗

To appreciate why citation analysis matters, it helps to see where traditional classification systems fall short. Schemes like the Dewey Decimal Classification (DDC) and the Library of Congress Classification (LCC) have organised libraries for over a century. They are excellent at one job: giving every book a single, fixed location on a shelf. But that strength is also their weakness when we try to map living, changing knowledge.

The single-location problem 🔗

Traditional schemes are built on rigid hierarchies, where each work must sit in exactly one place. A book on environmental economics has to be filed either under economics or under environmental science, even though it belongs to both. The physical arrangement forces a choice, and the connection between the two fields is lost on the shelf. This is fine for shelving but poor for representing how subjects actually relate.

Keeping pace with interdisciplinary knowledge 🔗

Knowledge is not static, and this is where rigid systems strain the most. New fields like data science, climate studies, and gender studies cut across several established categories at once. Standard classification schemes are known to suffer from limited granularity and a slow ability to adapt to emerging interdisciplinary fields, which can lead to misclassification. Because a hierarchy must anticipate categories in advance, genuinely new combinations of ideas have nowhere natural to go. The faceted, analytico-synthetic approach pioneered by S. R. Ranganathan was an important response to this rigidity, since it builds class numbers by combining facets rather than choosing from a fixed list – but even facet analysis describes a document, it does not draw the wider map of how whole fields connect.

How citation analysis offers a more holistic view 🔗

Citation analysis sidesteps the single-location problem entirely. A paper can belong to many clusters at once, linked to every other work it cites or shares references with. There is no shelf to limit it. This produces a relational, multidimensional picture in which interdisciplinary bridges show up clearly instead of being hidden. Where classification asks “which one box does this belong in?”, citation analysis asks “what is this connected to?” – and the second question fits the reality of modern knowledge far better. Importantly, the two approaches are complementary. Classification still organises collections for access, while citation mapping reveals the intellectual structure behind them.

Graphic displays of subject structures 🔗

Numbers in a citation matrix are hard to interpret. The real value of mapping appears when those relationships are turned into a picture. Graphic displays translate the mathematics of citation links into spatial maps that the human eye can read at a glance.

From matrices to maps 🔗

The process usually starts with a co-citation or co-occurrence matrix, which records how often items appear together. Statistical methods then convert these similarities into positions on a two-dimensional space. Multidimensional scaling places strongly related items close together and weakly related items far apart, while clustering groups them into coloured regions that represent sub-fields. The output is typically a network graph: each node is a document, author, journal, or keyword, and each line is a relationship. Larger nodes mean greater influence, and tight clusters mean a coherent specialty. A reader can immediately spot the core of a field, its branches, and the thin links that connect distant areas.

Tools that draw the map 🔗

Several free tools have made this kind of mapping widely accessible, which matters for students and researchers working with limited budgets. VOSviewer, developed by Nees Jan van Eck and Ludo Waltman, pays special attention to the graphical side and is designed to display large bibliometric maps in an easy-to-interpret way. CiteSpace, created by Chaomei Chen, takes a more analytical route and pinpoints research focal points by analysing clusters of publications over time. The Bibliometrix R package and its web app Biblioshiny offer a third option for building network diagrams. Many analysts pair them, using one for clear visuals and another for deeper trend analysis.

Reading a map with care 🔗

A science map is persuasive, but it is still a model, not the territory. Every map depends on choices: which database supplied the data, which time span was used, and which clustering settings were applied. Change these, and the picture can shift. A visualization also gives a simplified view of complex data, so it should be read alongside other evidence rather than treated as the final word. For a student, the lesson is to treat a knowledge map as a starting point for questions, not as a conclusion. Used this way, graphic displays of subject structures become one of the most powerful tools we have for understanding how knowledge is built, shared, and connected.

What do you think? If a single research paper can belong to many clusters at once, does the idea of a fixed “place” for knowledge still make sense in the digital age? And when a citation map and a traditional classification scheme disagree about where a subject belongs, which one would you trust more, and why?

References

1. https://www.science.org/doi/10.1126/science.122.3159.108
2. http://eprints.rclis.org/17524/1/Co-citation.pdf
3. https://www.ideals.illinois.edu/items/8226/bitstreams/28129/data.pdf
4. https://www.cwts.nl/tvr/documents/avr-cocit-word-i.pdf
5. https://pmc.ncbi.nlm.nih.gov/articles/PMC387295/
6. https://arxiv.org/pdf/2512.22524
7. https://link.springer.com/article/10.1007/s11192-009-0146-3
8. https://www.sciencedirect.com/science/article/pii/S2666166724004349