Homophily and missing links in citation networks

Citation networks have been widely used to study the evolution of science through the lenses of the underlying patterns of knowledge flows among academic papers, authors, research sub-fields, and scientific journals. Here we focus on citation networks to cast light on the salience of homophily, namely the principle that similarity breeds connection, for knowledge transfer between papers. To this end, we assess the degree to which citations tend to occur between papers that are concerned with seemingly related topics or research problems. Drawing on a large data set of articles published in the journals of the American Physical Society between 1893 and 2009, we propose a novel method for measuring the similarity between articles through the statistical validation of the overlap between their bibliographies. Results suggest that the probability of a citation made by one article to another is indeed an increasing function of the similarity between the two articles. Our study also enables us to uncover missing citations between pairs of highly related articles, and may thus help identify barriers to effective knowledge flows. By quantifying the proportion of missing citations, we conduct a comparative assessment of distinct journals and research sub-fields in terms of their ability to facilitate or impede the dissemination of knowledge. Findings indicate that knowledge transfer seems to be more effectively facilitated by journals of wide visibility, such as Physical Review Letters, than by lower-impact ones. Our study has important implications for authors, editors and reviewers of scientific journals, as well as public preprint repositories, as it provides a procedure for recommending relevant yet missing references and properly integrating bibliographies of papers.

💡 Research Summary

The paper investigates the role of homophily—the tendency of similar items to connect—in the formation of citation networks and introduces a method to detect missing citations between highly related papers. Using a comprehensive dataset of all articles published in the American Physical Society (APS) journals from 1893 to 2009, the authors develop a statistically rigorous similarity measure based on the overlap of reference lists.

Traditional similarity metrics such as the Jaccard index treat the overlap as a simple ratio and fail to account for differences in reference list size and the varying importance of cited works. To overcome these shortcomings, the authors adapt the Statistically Validated Network (SVN) framework, originally designed for bipartite user‑item systems, to directed citation graphs. They define two disjoint node sets: A, the set of citing papers, and B, the set of cited papers that have received at least two citations. For each in‑degree class k in B, they construct a homogeneous pool S_k of papers with exactly k citations and consider all citing papers that reference any member of S_k.

For any pair of citing papers i and j within the same S_k, the number of common references X follows a hypergeometric distribution under the null hypothesis of random citation selection. The p‑value q_{ij}(k) is the probability of observing at least the actual number of shared references N_{ij}. Because each pair can be evaluated across multiple k values, the authors apply a False Discovery Rate (FDR) correction to control for multiple testing. Pairs whose minimum corrected p‑value falls below a chosen significance threshold p* (e.g., 0.01) are deemed statistically similar.

Having identified statistically significant similarity pairs, the authors examine how citation probability depends on similarity. By binning similarity scores and computing the fraction of pairs that are actually linked, they demonstrate a clear monotonic increase: the higher the validated similarity, the greater the likelihood of a citation. This empirical finding confirms that homophily is a driving force in citation network growth.

The second major contribution is the systematic detection of “missing citations.” For pairs with high similarity (above a high percentile) that lack a directed link, the authors interpret the absence as a potential knowledge‑flow gap. They quantify the proportion of such missing links for each APS journal and for major physics sub‑fields. Results show that high‑visibility journals, especially Physical Review Letters, exhibit the lowest missing‑citation rates, whereas lower‑impact journals have substantially higher rates. Among sub‑fields, Electromagnetism and Interdisciplinary Physics have the smallest fractions of missing links, suggesting more cohesive citation practices, while other areas such as condensed‑matter or statistical physics display larger gaps.

These findings have practical implications. The validated similarity measure can be integrated into manuscript preparation tools, pre‑print servers, or editorial workflows to suggest relevant but uncited literature, helping authors build more complete bibliographies and reducing inadvertent omission of important prior work. Editors can use missing‑citation diagnostics to flag potential oversights during peer review, and funding agencies or policy makers can assess the efficiency of knowledge dissemination across disciplines and publication venues.

The authors acknowledge limitations. Relying solely on reference‑list overlap ignores textual content, keyword similarity, and semantic topic modeling, which could enrich the similarity assessment. Moreover, citations serve diverse purposes (e.g., self‑citation, negative citation, perfunctory citation) that are not distinguished in the current framework. Future work could combine bibliographic overlap with natural‑language processing, incorporate citation intent classification, and extend the methodology to other scientific domains beyond physics.

In summary, the study provides a robust statistical tool to measure article similarity, empirically validates homophily in citation formation, and offers a systematic approach to uncover missing citations, thereby contributing both to the theoretical understanding of scientific knowledge flows and to practical improvements in scholarly communication.