The Closer the Better: Similarity of Publication Pairs at Different Cocitation Levels

Abstract

We investigate the similarities of pairs of articles which are co-cited at the different cocitation levels of the journal, article, section, paragraph, sentence and bracket. Our results indicate that textual similarity, intellectual overlap (shared references), author overlap (shared authors), proximity in publication time all rise monotonically as the co-citation level gets lower (from journal to bracket). While the main gain in similarity happens when moving from journal to article co-citation, all level changes entail an increase in similarity, especially section to paragraph and paragraph to sentence/bracket levels. We compare results from four journals over the years 2010-2015: Cell, the European Journal of Operational Research, Physics Letters B and Research Policy, with consistent general outcomes and some interesting differences. Our findings motivate the use of granular co-citation information as defined by meaningful units of text, with implications for, among others, the elaboration of maps of science and the retrieval of scholarly literature. IntroductionThe co-citation relation is used extensively in bibliometrics, and has received some recent attention in information retrieval. Applications include the identification of topically-related publications for search engines and clustering of publications to understand the structure of science. If two or more publications are co-cited by a third one, they are generally assumed to be related to some extent, from the viewpoint of their citing authors (Small 1973). Normally, this assumption is considered to be valid already at a relatively coarse co-citation level, most often at the publication (e.g. article) level. In addition, recent work suggests that the relatedness of co-cited publications might increase with increasing proximity of two publications within the full text of the citing publication (e.g. Gipp & Beel 2009), and that improvements in maps of science or document retrieval can be obtained by taking textual proximity into account (e.g. Boyack et al. 2013). Indeed, it makes sense to assume that if two publications are co-cited within the same sentence or bracket in a publication, they typically will be in some way more related than two publications co-cited only at the more general section or publication levels. Yet open questions remain. We know little about the ways in which related, co-cited publications are similar over different dimensions. Furthermore, to what extent do different notions of similarity, such as textual and intellectual, depend on the level of the co-citation? This study was designed to provide answers to these questions.

Technical research innovations of the US national security system

“…In order to create the network of documents, links between documents were weighted using the intellectual overlap equation (Colavizza, Boyack, Eck, & Waltman, 2017), and selecting the Top-15 similarities with procedures proposed by Boyack and Klavans (2010).…”

Creativity in science and the link to cited references: Is the creative potential of papers reflected in their cited references?

“…Co-citation has been widely used since its introduction (Boyack & Klavans, 2010) which indicates its importance in the bibliometric community. For example, Colavizza, Boyack, van Eck, and Waltman (2017) investigated the similarity of publication pairs at different co-citation levels: journal, article, section, paragraph, sentence, and bracket (including the citations). They found that the similarity of publication pairs increases monotonically as the co-citation level gets lower (from journal to bracket).…”

Combining metadata and co-citations for recommending related papers

“…Among bibliographic-based approaches, co-citation is the oldest and most widely used approach. State-of-the-art approaches incorporate content with co-citation and suggest that the actual similarity between co-cited documents is related to the proximity of citations in the text [7,8,18,19]. However, the problem with state-of-the-art approaches again arises when the content is not available as the main ingredient for these approaches is missing.…”

“…Usually, the content contains details about motivation, methods, datasets, results, sections, citation context, multiple mentions of a citation, etc. Therefore, full-text options serve as a fundamental element of these approaches and are often provided in the form of key terms [7][8][9][10]. Key terms are words or phrases that are assumed to be more meaningful.…”

“…Content-independent approaches (such as collaborative filtering [11][12][13], graph-based [14], and hybrid approaches [7,8,15]) are also proposed in recommender systems. Collaborative filtering is based on the idea of "like-minded people like/dislike the same things".…”