Citation Graph
A network of academic papers connected by citation links — the data structure underlying tools like Connected Papers, Litmaps, and ResearchRabbit that visualize how papers relate to each other.
What it means
A citation graph is a network (graph) where each node is an academic paper and each directed edge represents a citation — pointing from the citing paper to the paper it cites. It’s the data structure that underlies academic search engines and literature discovery tools.
Key terms in citation graph analysis:
- Forward citations (cited by): Papers that cite your paper — useful for tracking how a piece of work has been used or extended
- Backward citations (references): Papers your paper cites — the direct intellectual lineage
- Citation depth / chains: Following citations recursively — your paper cites paper A, which cited paper B, which cited paper C — to trace an idea back to its origins
- Co-citation: Two papers that are frequently cited together, even if they don’t cite each other — a signal of conceptual relatedness
- Bibliographic coupling: Two papers that cite many of the same references, indicating they address the same topic from the same intellectual tradition
Why it matters for researchers
Literature discovery tools are built on citation graphs. When Connected Papers shows you papers related to your seed paper, it’s using co-citation similarity — papers that are cited together often are near each other in the graph. Litmaps displays the graph chronologically. ResearchRabbit traverses citation chains to suggest papers you haven’t read.
Citation counts are derived from the graph. The number of times a paper has been cited is simply the in-degree of its node in the citation graph. Different databases (Scopus, Semantic Scholar, OpenAlex) build their own citation graphs from different source sets, which is why citation counts vary across platforms.
The h-index and other bibliometric measures are computed over the citation graph. Understanding that these are graph-theoretic measures helps interpret their limitations — they measure how the graph is structured, not the quality of the underlying work.