Co-Citation Analysis

The core metric: a raw count of how many later documents cite both Source A and Source B together. A high frequency suggests a strong semantic relationship or shared evidentiary context.

• Example: If 150 papers cite both 'Attention Is All You Need' and 'BERT: Pre-training', their co-citation frequency is 150.
• This metric is dynamic; it increases as new literature integrates both sources.

Co-citation counts are used to construct a similarity matrix, which is then fed into clustering algorithms. This groups documents into research fronts or specialty areas.

• Process: Normalize co-citation counts -> Generate a proximity matrix -> Apply hierarchical or k-means clustering.
• Outcome: A map of science where nodes are papers and edges are co-citation links, revealing schools of thought invisible to keyword search.

A variant that maps the intellectual structure of a discipline by analyzing which authors are frequently cited together, rather than individual papers.

• Purpose: Identifies invisible colleges and schools of thought by grouping authors with shared theoretical or methodological approaches.
• Example: In computer science, a cluster might form around authors like Geoffrey Hinton, Yann LeCun, and Yoshua Bengio, visually representing the deep learning research community.

Co-citation strength is not static. Tracking it over time reveals the evolution of ideas and the rise and fall of research paradigms.

• Citation Burst Detection: Identifies sources that experience a sudden surge in co-citations, signaling an emerging trend or breakthrough.
• Obsolescence: A declining co-citation rate can indicate that a source's findings have become common knowledge or been superseded, informing source recency weight calculations.

Once a co-citation network is built, graph theory metrics identify the most influential nodes.

• Betweenness Centrality: Measures how often a document sits on the shortest path between two other documents. High betweenness indicates a gatekeeper or interdisciplinary bridge.
• Eigenvector Centrality: Measures a node's influence based on the influence of its neighbors. A document co-cited with other highly co-cited documents receives a higher score, analogous to Citation Graph Rank.

Co-citation analysis is a retrospective measure, looking backward from the citing papers. In contrast, bibliographic coupling is a prospective measure, linking two documents that share common references in their own bibliographies.

• Co-citation: 'Paper C cites A and B' -> A and B are linked. This captures how later literature synthesizes prior work.
• Bibliographic Coupling: 'Paper A and Paper B both cite Paper X' -> A and B are linked. This captures foundational similarity at the time of publication.

A complementary metric to co-citation that measures similarity between two sources based on the number of references they share. While co-citation looks forward at who cites them together, bibliographic coupling looks backward at their shared intellectual foundation.

• Two papers citing the same 15 foundational works have high coupling strength
• Used to identify corroborating evidence and thematic clusters
• Strong coupling suggests shared methodology or theoretical grounding

An algorithmic assessment of a source's importance within a citation network, analogous to PageRank. Authority is derived from the quantity and quality of inbound links from other credible sources.

• A source cited by many high-authority papers receives an elevated rank
• Co-citation clusters often reveal which nodes are authoritative hubs
• Forms the backbone of systems like Google Scholar's relevance sorting

A verification technique that checks for agreement among multiple independent, high-quality sources to confirm a claim. Co-citation analysis identifies which sources are frequently grouped together, enabling consensus scoring.

• If five independently authored papers co-cite the same two sources for a claim, confidence increases
• Reduces reliance on any single potentially biased source
• Critical for hallucination prevention in RAG systems

A high-dimensional embedding that mathematically represents the contextual meaning of a source document. While co-citation measures structural relationships, semantic relevancy vectors measure topical alignment.

• Computed using transformer models to encode document meaning
• Combined with co-citation data to distinguish between thematic and incidental co-occurrence
• Enables precise claim-source alignment scoring

A dynamic, interconnected model representing entities (authors, institutions, domains) and their trust relationships. Co-citation patterns feed into these graphs as weighted edges between source nodes.

• Propagates authority scores across the network
• A domain frequently co-cited with .edu institutions inherits trust signals
• Used to compute Authoritative Domain Boost and detect citation cartels

A measure of the completeness and logical validity of the path from an AI's output claim back through its citations to foundational, verifiable data. Co-citation analysis strengthens chain integrity by revealing whether cited sources are part of a coherent, well-connected research tradition.

• Fragmented or isolated co-citation clusters may indicate weak evidence chains
• High integrity requires recursive validation through primary sources
• Essential for auditability in regulated industries