[Paper Review] Are self-citations a normal feature of knowledge accumulation?

Philippe Vincent‐Lamarre, Vincent Larivière|arXiv (Cornell University)|Mar 5, 2023

scientometrics and bibliometrics research4 citations

TL;DR

This study analyzes over 63 million articles and 51 million authors to examine self-citations in scholarly communication, finding that self-citations are more common among men and older researchers, moderately boost citation rates overall, and are more relevant than other citation types—though they significantly inflate metrics for a subset of researchers. The paper concludes that self-citations are a normal but strategically influential feature of knowledge accumulation.

ABSTRACT

Science is a cumulative activity, which can manifest itself through the act of citing. Citations are also central to research evaluation, thus creating incentives for researchers to cite their own work. Using a dataset containing more than 63 million articles and 51 million disambiguated authors, this paper examines the relative importance of self-citations and self-references in the scholarly communication landscape, their relationship with the age and gender of authors, as well as their effects on various research evaluation indicators. Results show that self-citations and self-references evolve in different directions throughout researchers' careers, and that men and older researchers are more likely to self-cite. Although self-citations have, on average, a small to moderate effect on author's citation rates, they highly inflate citations for a subset of researchers. Comparison of the abstracts of cited and citing papers to assess the relatedness of different types of citations shows that self-citations are more similar to each other than other types of citations, and therefore more relevant. However, researchers that self-reference more tend to include less relevant citations. The paper concludes with a discussion of the role of self-citations in scholarly communication.

Motivation & Objective

• To investigate the role of self-citations in knowledge accumulation across academic careers.
• To analyze how self-citations relate to author demographics, including gender and age.
• To assess the impact of self-citations on research evaluation indicators such as citation rates.
• To compare the topical similarity between cited and citing papers across citation types, especially self-citations.
• To evaluate whether self-citations contribute meaningfully to scholarly relevance or serve as metric manipulation tools.

Proposed method

• The study uses a dataset of 63 million articles and 51 million disambiguated authors from scholarly publication records.
• It distinguishes between self-citations (citations to one’s own works) and self-references (references to one’s own works in the text) using metadata and text analysis.
• Citation similarity is measured by comparing abstracts of citing and cited papers using NLP-based embedding techniques.
• Statistical models assess the relationship between self-citation frequency and author demographics (gender, career stage).
• The impact of self-citations on citation rates is quantified using regression analysis, controlling for field and publication age.
• The study compares the topical relatedness of self-citations versus other citation types using cosine similarity of abstract embeddings.

Experimental results

Research questions

• RQ1 How do self-citations and self-references evolve across researchers’ academic careers?
• RQ2 Are self-citations more prevalent among men or older researchers?
• RQ3 To what extent do self-citations inflate citation-based research evaluation metrics?
• RQ4 How similar are the topics of self-cited papers compared to other citation types?
• RQ5 Do researchers who self-reference more tend to include less relevant citations?

Key findings

• Self-citations are more common among men and older researchers, with a significant increase in frequency as researchers age.
• On average, self-citations have a small to moderate positive effect on authors’ overall citation rates.
• A subset of researchers experience highly inflated citation counts due to self-citations, indicating disproportionate impact on metrics.
• Self-citations are more topically similar to each other than other citation types, indicating higher relevance.
• Researchers who self-reference more tend to include less relevant citations, suggesting a trade-off between self-promotion and citation quality.
• The study finds that self-citations are not merely redundant but are often topically coherent, supporting their role in knowledge continuity.