Source-linked AI summary
Visualizing the context of citations referencing papers published by Eugene Garfield: A new type of keyword co-occurrence analysis
Lutz Bornmann, Robin Haunschild, Sven E. Hug
TL;DR
Citation-context analysis faces substantial effort in producing its data. This study builds keyword co-occurrence networks from contexts around citations to Eugene Garfield’s papers and compares them with title-and-abstract keyword networks. The comparison suggests that citation contexts are semantically closer to Garfield’s papers than to the papers citing them.
Problem
Producing citation-context data requires substantial effort to delimit, extract, and categorize the text around citations.
Source: §2 · p. 4Method
The study builds a keyword co-occurrence network from citation contexts around references to Garfield’s papers and compares it with title-and-abstract keyword networks.
Source: §1 · p. 3, §5 · p. 15Results
Garfield’s papers and citation contexts from papers citing them are semantically more closely related than citation contexts and citing papers’ title-and-abstract keywords.
Source: §5 · p. 16Takeaways & Limitations
The comparison suggests that citation contexts align more closely with descriptions of cited papers than with descriptions of citing papers.
Source: §5 · p. 16The study explores a new approach to citation contexts rather than providing a complete description of Garfield’s work.
Source: §4 · p. 8
Abstract
from arXiv · showhide
During Eugene Garfield's (EG's) lengthy career as information scientist, he published about 1,500 papers. In this study, we use the impressive oeuvre of EG to introduce a new type of bibliometric networks: keyword co-occurrences networks based on the context of citations, which are referenced in a certain paper set (here: the papers published by EG). The citation context is defined by the words which are located around a specific citation. We retrieved the citation context from Microsoft Academic. To interpret and compare the results of the new network type, we generated two further networks: co-occurrence networks which are based on title and abstract keywords from (1) EG's papers and (2) the papers citing EG's publications. The comparison of the three networks suggests that papers of EG and citation contexts of papers citing EG are semantically more closely related to each other than to titles and abstracts of papers citing EG. This result accords with the use of citations in research evaluation that is based on the premise that citations reflect the cognitive influence of the cited on the citing publication.
1 Introduction
The study introduces keyword co-occurrence networks built from citation contexts around references to Garfield’s papers, a source not previously used for this purpose. It compares this approach with conventional title-and-abstract keyword networks.
1 Introduction
- Garfield founded ISI, which provided citation indexes that preceded Web of Science, and developed an early bibliometric network with Sher and Torpie.The work was based on citations as recursive operations and ultimately led to HistCite.Source: §1 · p. 3
- The study builds keyword co-occurrence networks from words located around citations to papers in Garfield’s publication set.A citation context is defined as the words around a specific citation.Source: §1 · p. 3
- To the authors’ knowledge, citation context had not previously been used as a data source for generating keyword co-occurrence networks.Earlier keyword co-occurrence networks generally used words from titles, abstracts, or author-generated keyword lists.Source: §1 · pp. 3–4
2 Literature overview and research questions
The section describes the effort involved in producing citation-context data, notes that Microsoft Academic made segmented contexts downloadable, and sets out comparisons among three keyword networks.
2 Literature overview and research questions
- Producing citation-context data requires substantial effort to delimit, extract, and categorize the text around citations, and past studies were often small-scale.Source: §2 · p. 4
- Microsoft Academic was presented as a large-scale scholarly database that allowed citation contexts, already segmented, to be downloaded.Source: §2 · p. 5
- The study compares a citation-context network with title-and-abstract keyword networks based on Garfield’s papers and papers citing them.Source: §2 · p. 5
- The authors expect the citation-context network to resemble the network from cited papers more than the one from citing papers, reflecting their stated view of citation context.They frame the comparison around the premise that citations reflect cognitive influence from cited to citing publications.Source: §2 · p. 5
3 Methods
The study uses Microsoft Academic data to construct three keyword co-occurrence maps: one each from Garfield’s papers, papers citing them, and citation contexts around those citations.
3.1 Microsoft Academics (MA)
- Microsoft Academic data were retrieved through its Academic Knowledge API, a database described as rapidly expanding and updated weekly.The database grew from 83 million records in 2015 to 140 million in 2016 and 168 million in early 2017.Source: §3.1 · pp. 5–6, §3.2 · p. 6
3.2 Dataset used
- Microsoft Academic contained 327 of Garfield’s 1,558 publications; citation contexts were available for 59 papers, with 428 contexts extracted.The researchers retrieved title and abstract information for the 59 papers and 343 citing papers from Web of Science using DOI searches.Source: §3.2 · p. 6
3.3 Keyword co-occurrence analysis
- The researchers produced three maps: title-and-abstract keywords from Garfield’s papers, those from papers citing them, and keywords from citation contexts.Source: §3.3 · p. 6
- VOSviewer extracted adjectives and nouns, treating keywords in the same title, abstract, or citation context as co-occurring; closer map positions indicate stronger co-occurrence.All maps used binary counting, a minimum occurrence of four, and the 60% most relevant keywords; irrelevant terms were manually excluded.Source: §3.3 · p. 7
4 Results
The title-and-abstract network for papers citing Garfield was broader than the network for Garfield’s papers, while citation contexts yielded a smaller set of clusters with both shared and distinct topics. The results are based on a restricted set of Garfield publications and capture only part of his influence.
4 Results
- The networks reflect a restricted publication set; the citing papers and contexts represent only part of Garfield’s influence or impact.Although 327 of Garfield’s 1,558 publications were found in Microsoft Academic, citation contexts were available for only 59.Source: §4 · p. 8
- Garfield’s title-and-abstract map used 15 keywords, and its two clusters were fewer than the 18 identified for citing papers.The 607 extracted keywords included 27 occurring at least four times; 15 met the inclusion criteria.Source: §4 · pp. 7, 9, Figure 1 · p. 8
- Titles and abstracts of papers citing the 59 Garfield publications yielded 6,878 keywords, with 297 retained for the map.Of 6,878 extracted keywords, 512 occurred at least four times; 297 met the inclusion criteria.Source: §4 · p. 9
- The citing-paper map contained 18 clusters, compared with two for Garfield’s papers; citation contexts yielded three clusters.The citation-context clusters were Journal Impact Factor, historiographs, and Garfield as founder of the ISI.Source: Figure 1 · p. 8, Figure 2 · p. 9, §4 · pp. 9, 11
- The 18 citing-paper clusters span bibliometric indicators, data providers, analytical units, citation types, methods, and research evaluation.The reported topics include the Journal Impact Factor, h-index, nations, self-citation, citation networks, and evaluation.Source: §4 · pp. 9–11
- The citation-context map retained 97 keywords from 2,347 extracted keywords and identified Journal Impact Factor, historiographs, and Garfield as founder of the ISI.Of 2,347 extracted keywords, 184 occurred at least four times; 97 met the inclusion criteria.Source: §4 · p. 11, Figure 3 · p. 12
- The Journal Impact Factor cluster in Garfield’s papers focuses on the indicator itself; citing-paper keywords connect it to alternatives, while normalization appears in both citing-paper and context networks.The relationship between the Journal Impact Factor and alternative indicators was not found in citation-context keywords.Source: §4 · p. 13
5 Discussion
The discussion presents citation-context keyword co-occurrence networks as a new way to visualize how cited studies are perceived. Comparing three networks, it finds citation contexts more semantically aligned with EG’s papers than with citing papers’ titles and abstracts, while noting limits in the available data.
5 Discussion
- The study introduces keyword co-occurrence networks built from words extracted around citations to visualize citation-context data.The authors describe this as the first study to produce such networks and focus primarily on the approach’s opportunities, secondarily on its results.Source: §5 · p. 15
- Citing-paper title and abstract networks cover broad topics, but their overlap with the cited-paper and citation-context networks is marginal, with no historiography keyword identified.The discussion also reports that no cluster referred to historiography.Source: §5 · p. 15
- The citation-context network associates JIF with SNIP, percentiles, and the h-index, JCR with JIF scores and the Eigenfactor, and SCI with patents.The three clusters representing EG’s achievements have different keyword associations in the citation-context network than in the other two networks.Source: §5 · p. 15
- EG’s papers and citation contexts from papers citing EG are semantically more closely related than citing papers’ titles and abstracts.The comparison supports the expectation that citation-context keywords relate more closely to the cited than the citing publications.Source: §5 · p. 16
- The analysis is limited by a restricted set of EG papers in Microsoft Academic and unavailable citation contexts for many citing papers.The authors recommend larger datasets spanning broad subject areas for future comparisons of cited papers, citing papers, and citation contexts.Source: §5 · p. 16