When we look at a scientific paper's bibliography, we are looking at a map of its intellectual ancestry. Some papers draw heavily on foundational work written decades ago, while others cite almost exclusively from the recent past. This temporal dimension of science—how far back a field looks, and how quickly its past ideas fade from view—is the focus of temporal citation lifecycle analysis. By tracking metrics like the mean citation age, researchers can uncover a discipline's "knowledge metabolism," revealing whether a field is building upon a stable theoretical core or suffering from collective amnesia.
Measuring the Age of Knowledge
How do we quantify a field's historical awareness? The primary tool is the mean age of citation, which calculates the average time elapsed between the publication of a citing paper and the works it references. As explained by Janvijay Singh et al. in Forgotten Knowledge: Examining the Citational Amnesia in NLP, this metric serves as a proxy for the historical depth of a field's foundational context.
To compare these temporal spans across different eras without being skewed by the sheer explosion of modern publishing volume, researchers use probabilistic measures. For instance, in an analysis of the ACL Anthology, Marcel Bollmann et al. utilized the Common Language Effect Size (CLES) in On Forgetting to Cite Older Papers: An Analysis of the ACL Anthology to determine the likelihood that a randomly selected citation from one period is older than one from another.
Similarly, the authors of Review articles as windows into Knowledge accumulation: the case of AI research track "reference lag"—the temporal distance between an original article and the review articles that synthesize it—to measure the speed of knowledge obsolescence and foundational stability. When this temporal gap increases over time, as observed by Dongyu Zhang et al. in Metaphor Research in the 21st Century: A Bibliographic Analysis, it signals that a field has established a stable, classic core that contemporary researchers continue to rely upon.
Disciplinary Signatures and Knowledge Metabolism
Not all academic fields consume information at the same rate; every discipline possesses a unique "knowledge metabolism." As demonstrated by Yasuhiro Yamashita et al. in Tendency analysis of article citations in Japanese academic disciplines, hard sciences typically reach peak citation influence within two to three years before experiencing rapid decay. In contrast, the humanities and social sciences exhibit a much slower metabolism, sustaining their impact over much longer periods. Because of these variations, standardized evaluation windows fail to fairly capture research impact across diverse domains.
These metabolic differences become especially visible when distinct fields interact. Rima Hazra et al. note in The Rise and Rise of Interdisciplinary Research: Understanding the Interaction Dynamics of Three Major Fields - Physics, Mathematics and Computer Science that identifying these temporal signatures is necessary to understand the pace of cross-disciplinary innovation.
In a follow-up study modeling these interactions, Rima Hazra et al. revealed in Modeling interdisciplinary interactions among physics, mathematics and computer science that these relationships are often highly asymmetric. For example, computer science papers tend to cite contemporary internal research but look far back into the past for mathematical foundations. Conversely, mathematics papers tend to cite older internal literature but eagerly leverage highly recent, cutting-edge computer science tools.
The Digital Paradox: Access vs. Amnesia
Intuitively, one might expect that digitizing the world's literature would make older papers easier to find, thereby increasing the average age of citations. However, empirical reality reveals a striking paradox.
On one hand, some evidence suggests digitization deepens our historical reach. Yuxiao Dong et al. found in A Century of Science: Globalization of Scientific Collaborations, Citations, and Innovations that the average age of cited literature nearly doubled over a century, suggesting that digital archives successfully reduced the friction of discovering historical knowledge. Furthermore, Xianwen Wang et al. showed in The open access advantage considering citation, article usage and social media attention that open-access mechanisms flatten the decay curve, allowing papers to maintain a "long tail" of sustained discovery and usage over time.
On the other hand, a powerful counter-trend exists: recency bias. In a landmark study, James Evans discovered in Publikacje elektroniczne a zwężenie obszaru poszukiwań nauki i wiedzy that as journals moved online, the actual breadth of cited research began to shrink. Because modern search interfaces and digital platforms prioritize recent and highly relevant content, researchers hyper-focus on a narrow band of recent work, causing older foundational literature to be bypassed.
This "citational amnesia" is particularly acute in fast-moving fields. Jan Philip Wahle et al. analyzed 240 million articles in Citation Amnesia: On The Recency Bias of NLP and Other Academic Fields and confirmed a growing recency bias across academia. They found that machine learning and natural language processing lead this trend, frequently ignoring foundational roots in favor of the latest state-of-the-art benchmarks.
Modeling the Decay of Attraction
To understand why these patterns emerge structurally, network scientists model citations not just as static graphs, but as evolving systems governed by aging effects. Traditional network models assume that popular papers simply get richer over time through preferential attachment. However, real-world data shows that papers eventually lose their ability to attract attention.
To capture this reality, Xiu-Xiu Zhan et al. introduced evolutionary models using hypergraph theory in The Aging Effect in Evolving Scientific Citation Networks. By incorporating a collective "attraction decay" parameter, the authors accurately replicated the real-world aging effects observed in physics and computer science citation networks. This mathematical framing proves that knowledge turnover is a measurable, fundamental trait of scientific networks, directly shaping how long an idea remains active before fading into the historical bedrock of science.
Key Foundational Work
Publikacje elektroniczne a zwężenie obszaru poszukiwań nauki i wiedzy, przeł. Tomasz Olszewski (2008) — contributed 3 concept elements to this direction.
Tendency analysis of article citations in Japanese academic disciplines (2000) — contributed 3 concept elements to this direction.